The findings are thrilling. The findings are null. The findings are promising. Or maybe disappointing… These findings can be whatever you want them to be. There is something there for everybody.

Open Dialogue has generated a lot of enthusiasm since it emerged in Finland in the 80s as a novel approach to treating psychosis. It racked up some evidence over the year as it spread internationally, but had never been subjected to a randomised controlled trial. Ten years ago, UK-based researchers embarked on an ambitious project to do just that. So after such a long wait, it’s a big deal that the ODDESSI RCT is finally out.

The results have fuelled media excitement. They have sparked academic critique. They have drawn praise, and calls for expanded provision. Their dissemination received a rebuke from the very journal that published the trial.

In short, the primary outcome — time to relapse — showed no significant difference compared with treatment as usual. But there were enough significant, and striking, results amongst the secondary outcomes (e.g. lower rate of hospitalisation and greater self-reported satisfaction with services) for OD advocates to celebrate it as a success. However, RCT ethos denotes that primary outcomes should be just that. As such, the emphasis placed on secondary outcomes that favour OD, over the primary outcome that does not, has antagonised some researchers.

There has been intense debate about what these results represent, how they should be interpreted, and what they mean for the future of OD in the UK, and indeed the future of NHS mental health services more broadly.

Yet the vigour of these debates is perhaps unsurprising, because Open Dialogue seems to have a particular capacity to galvanise and inspire clinicians on the one hand, and to polarise and repel on the other.

Some friction was already notable during the conduct of the RCT, according to an ethnographic study by Chase and Mosse. They suggest the OD practitioner advocates and the trial researchers embodied two different moral philosophies. The researchers were situated within a utilitarian philosophy, which judges actions based on their consequences, and seeks to maximise positive outcomes, in this case through the tools of evidence-based medicine. The practitioner advocates, on the other hand, were situated within a “virtue ethics” framework, in which actions are judged first and foremost by the virtues they embody. From this perspective, the results of the trial mattered less than the inherent virtue of OD, as summed up by a quote from one of the interviewees:

“Morally and ethically it’s quite hard to argue against the principles of it. So, what exactly are we worried about? Why is there even the ODDESSI trial? Because should we not just be doing this stuff?” (participant, Chase & Mosse, 2025, p.644)

These contrasting philosophical positions go some way to explaining tensions around the trial. But I think this misses something important — the role of identity and group dynamics.

As such, beyond all the discussion about primary vs secondary outcomes etc., what caught my attention was a post by Mark New, a support worker in a service that almost participated in the ODDESSI trial. Mark underwent the OD training — a one-week residential retreat — and describes his misgivings about the atmosphere of intense fervour and lack of space for questioning, or for critical or alternative perspectives.

Mark’s reflections resonate strikingly with a systematic review I spent the last few years working on. I thought I’d offer a snapshot of the findings, as they might illuminate the curious controversy that has swirled around this RCT.

TL;DR: Open Dialogue training and practice often forges cohesive, values-driven and committed groups of practitioners with a strong, shared social identity. This can be beneficial in many ways. However, when this group identity is strong, homogenous and ideologically charged, damaging intergroup tensions and competition can emerge.

What is OD?
If you want a full lowdown on Open Dialogue, there’s a good overview of the key principles here.

In short, it has two key components:

1. A particular relational stance in the moment with service users

2. An overall model for delivering mental health services

The relational stance is characterised by tolerating uncertainty (not rushing to decisions about treatment), valuing all voices, and levelling hierarchies (between professionals, and between professionals, service users and their social networks). In other words, professional “expertise” is not invested with more power and importance than the voice of lived experience.

In the service of this, Open Dialogue is steeped in the philosophy of Mikhail Bakhtin, a Russian literary critic who developed the concept of polyphony (multi-voicedness) through his analysis of Dostoyevsky’s novels. This concept of polyphony, and valuing multiple different perspectives, is one I will return to.

Open Dialogue implementation issues

Reading through studies looking at Open Dialogue implementation, I was struck by how organisations often encountered difficulties and conflicts, which sometimes proved insurmountable. These issues often revolved around identity dynamics, and the extent to which staff developed identities compatible with OD.

So, myself and colleagues conducted a systematic review and meta-ethnography (a method for synthesising and interpreting qualitative papers) looking at clinician experiences of identity change when implementing or training in Open Dialogue. The full paper, which included 23 studies, is available open access here.

Through the synthesis, we derived four overarching themes and 11 subthemes, as you can see in the figure below.

Personal identity: the “satisfaction of being yourself”

Broadly, many clinicians experienced OD as an immensely rewarding way of working.

Particularly striking was the way OD practitioners described connecting with a ‘personal self’, and shedding their purely professional identity. There was a sense that OD training and practice allowed them to be themselves, or to “be authentic”. This coherence was presented as a contrast to the way they usually separated their personal and professional selves, and felt inauthentic at work.

“it’s like we actually get to do what we went to school for you know, or what our passions were” (participant, Uehling et al., 2024, p.4)

This was driven by a sense that OD principles aligned with their personal values, described as “philosophical fit” in one paper. OD was often perceived as more than just learning and deploying new skills and knowledge, but rather as a new “way of life” or “way of being”, that spilled over into their personal lives.

This had potential implications for wellbeing and burnout:

“I think a lot of the reason why I would get burned out was just being involved in things that didn’t feel good, that were inhumane, or that were taking people’s freedom away, or autonomy away and just watching the system be so cruel to the people that I am getting paid to help, year after year, it’s just horrible. And this work is probably the most humane way of work that I’ve ever done.” (participant, Florence et al., 2020, p.687)

Indeed, in the ODDESSI results presentation, Professor Russell Razzaque, a consultant psychiatrist and joint last author on the paper, said that, in his service in Havering, the treatment as usual team saw 100% staff turnover between 2017 and 2023. In the OD team, staff turnover was just 20%1. “When you go through this training, it creates a really cohesive team,” he explained in the presentation.

Shared OD group identity

The identity processes that unfolded within and between individuals contributed to the development of a shared Open Dialogue group identity.

Social identity theory provides a useful theoretical lens here. It offers a framework for understanding how people’s perception of group membership shapes their sense of self. When a social identity is salient in a given context, individual identity is subsumed within it. The individual perceives themselves more as a member of a group than as an individual. This can contribute to group cohesion and co-operation, but also intergroup conflict.

The development of this shared group identity was catalysed by several factors that are particular to OD:

1. The encouragement to share intimate personal stories nurtured close bonds between practitioners.

2. Intense residential training courses provided the conditions for rapid identity formation

3. Letting go of pre-existing professional identities (i.e. no longer identifying first and foremost as a psychiatrist or a mental health nurse, but rather as an OD practitioner).

4. The flattening of hierarchies dissolved the sense of difference and power differentials between practitioners, and fostered togetherness2.

Organisational research has linked social identification with numerous beneficial outcomes, including staff well-being, and motivation and well-being amongst care workers. These benefits were apparent in many of the studies included in the review.

Intergroup polarisation

But although the shared OD group identity could nurture cohesion and energise practitioners, it could also lead to challenges.

In some studies, the OD group identity was strong, homogenous and exclusionary. This was particularly notable when implementation was imposed by management from the top down, or when the OD group identity was ideologically charged. And sometimes this ideological charge was intense.

The sub-theme “Othering from the dominant paradigm”, captures how professionals aligned themselves with OD as ethically superior. This involved othering themselves from, and denigrating, the perceived dominant medical paradigm.

The “superior” part of this othering is important: this was not a case of saying, “here is an alternative”, but rather, “here is the alternative”. Implicit in this was an attack on people who don’t align themselves with OD.

“I think that there is a danger that people who practice OD see themselves as superior and this can be experienced as a provocation… as if in their infinite wisdom they know it all or they are better than other staff” (participant, von Peter et al., 2022, p. 3).

“[name of colleague] was there almost like a guru, a completely different person as far as I knew before. Hallelujah, how wonderful OD was and how we have all been doing it wrong so far” (participant, Lorenz-Artz et al., 2023, p. 8).

This is a dynamic that Mark New also mentions experiencing during his training:

“I felt a real case of ‘if you’re not with us, you’re against us’ emerging. Not merely a rejection of traditional healthcare, but a denouncement.”

In 11 papers, these kinds of ideological polarities provided fertile terrain for intergroup tensions.

Such processes have long been examined in social identity theory research. Outgroups are often perceived as threatening, particularly when they are represented as strong entities. This dynamic seems particularly likely to emerge when OD is framed as an underdog pushing back against a dominant, and therefore threatening, paradigm.

When important aspects of one’s self-concept are related to a group identity, one is motivated to enhance that ingroup identity, and draw favourable comparisons with outgroups. This social identity lens perhaps partly explains the allure of secondary outcomes that show superiority to TAU, over a primary outcome that does not.

Subscribe now

Intergroup collaboration

Intergroup tensions were not always inevitable or intractable. Across seven papers these intergroup polarities were softened, dissolved, or avoided in the first place.

Instead of intergroup polarisation, intergroup collaboration was more likely to occur when:

1. The OD group identity was heterogeneous, and made space for difference, nuance and complexity; in short, when the group identity was polyphonic

2. When implementation was bottom up, or top down and collaborative

3. When the OD group was inclusive, with permeable boundaries

Some services did manage to attain this intergroup collaboration, although it was far from the norm. This raises important questions about how the benefits of social identification—group cohesion, collaboration, self-esteem and sense of purpose—can be harnessed, without unleashing problematic intergroup conflicts. Organisational interventions derived from social identity theory hold some promise. But also, embedded within Open Dialogue philosophy is a potential solution.

Polyphony is a core OD principle. As such, it is a tragic irony when OD implementation is characterised and derailed by its opposite—monologism—as highlighted by Florence and colleagues. Welcoming polyphony may be crucial in avoiding the polarisation that has at times hampered OD implementation, and has emerged around the ODDESSI dissemination.

As Bakhtin writes:

“Truth is not born nor is it to be found inside the head of an individual person, it is born between people collectively searching for truth, in the process of their dialogic interaction.” (Bakhtin, 1984, Problems of Dostoyevsky’s Poetics, p.110)

This dialogic interaction ought to occur not just within OD circles, but between OD and the wider community of researchers and practitioners.

Conflicts of interest to declare: I love Bakhtin.

Subscribe now

Therapists, on average, do not improve with experience alone. Their outcomes stagnate or deteriorate. But they do not all follow the same trajectory. Some therapists do improve over time. Some deteriorate dramatically. Many stay at the same level they reached part way through their training.

Various factors contribute to this phenomenon. They can broadly be categorised as: 

1. The nature of therapy;

2. The nature of therapists;

3. The nature of humans.

Each one is in some ways conducive to the gradual erosion of therapist performance over time. But each of these factors also offers clues as to how experience can translate into improved, rather than gradually deteriorating, outcomes. In this post I’ll lay out each of these factors and the research that supports them. In the next post, I’ll explore some possible solutions.

The nature of therapy

Therapy unfolds in a noisy environment in which the relationship between the therapist’s actions and client outcomes is hard to intuit (Tracey et al., 2014).

As such, therapy is not the sort of endeavour that is readily conducive to the acquisition of expertise, which requires timely, accurate feedback in response to one’s actions. A basketball player practicing free throws can immediately and somewhat objectively perceive the effects of an adjustment to their foot position, and know whether it is an improvement to retain. In a few hours they can gather a sample of hundreds of outcomes on which to base this, whilst holding other variables steady. For a therapist, feedback is less immediate (typically weekly at best), and any adjustment they make to their practice is one independent variable among innumerable.

So many factors influence client outcomes beyond the therapist, including client factors (age, motivation etc.) and extratherapeutic factors (events in the client’s life). Did the therapist’s particular focus on accurate empathy contribute to the client’s recovery, or did the client just feel better because they got a new job or the Strait of Hormuz reopened? Did the therapist’s implementation of the Compassion Focused Therapy skills they learned during a recent workshop make the difference for this client, or would the client have simply got better anyway, due to the natural cycle of their mood? In fact, research shows therapist factors (e.g. empathic ability and facilitative interpersonal skills) account for only ~5% of variance in client outcomes (Johns et al., 2019).

There are complex and unpredictable interactions between the therapist’s actions and the unique individual client. What works with one person may not with another. In other words, anything “learned” from working with one client may not be relevant to the next one, which makes it hard to accumulate expertise across clients, although this may occur in very specific circumstances. Leon and colleagues (2005), for example, found that, when clients on a therapist’s caseload were matched in pairs based on demographic and clinical characteristics, the second client in the pair attained better outcomes, but only if they started therapy less than 75 days after the first. This very tentatively suggests some evidence of accumulated expertise, but only between similar clients, and only when they are treated in very close temporal proximity. These conditions are often not met in clinical practice.

Indeed, “outcomes” in themselves are hard to intuit, and therapists are surprisingly poor at perceiving and predicting them from their subjective judgement alone. This may sound implausible, given therapists’ reputation for attunement, but without tracking outcomes, they often struggle to infer their client’s trajectory of change.

In one study, 49 therapists were asked to predict which of their clients (n=550) would deteriorate at the end of treatment (Hannan et al., 2005). They were told that the deterioration base rate is roughly 8%. The sample predicted that 0.1% of their clients (3/550) would deteriorate, compared with 7.2% (40/550) who did.

In another study, Østergård and colleagues (2024), asked therapists to predict, after each session, whether their clients’ CORE-10 score would improve/ deteriorate/ stay the same by the end of therapy. Before making predictions, participants were informed that 5-14% of clients typically deteriorate during treatment. The participants made 1140 predictions, out of which they made only eight (0.7%) predictions of deterioration. All of these were made for clients who actually showed no change. Not a single one of the clients who did deteriorate was predicted to do so. In predicting deterioration, the therapists performed no better than chance.

More remarkable still, after each session, therapists were asked whether they believed their client had improved/ deteriorated/ stayed the same (on the CORE-10) based on what they had seen during the session. The therapists still performed no better than chance. Clients showed deterioration in 83/1015 sessions. Therapists identified 6 (7.2%) of these accurately. They misidentified 50 (60.2%) as “no change”, and misidentified 27 (32.5%) as improved. A tipsy aardvark with limited psychotherapy training, reclining by a pool, blindfolded by cucumber slices and guessing with his tongue could have performed similarly.

Without outcome measures, therapeutic change is clearly hard to accurately infer. And if therapists can’t accurately infer change, how can they adjust suitably in response? Unhelpful adjustments may be reinforced by inaccurate inferences about their effects.

Notably, these inaccurate inferences are all one way. The studies above demonstrate a systematic overoptimism bias—an erroneous belief that barely any clients will deteriorate, and indeed that clients have not deteriorated, even when they have.

Yet despite all these challenges, all this uncertainty, and all the limitations of their intuition, therapists are often very confident about the effects of their interventions.

The nature of humans

But what’s wrong with confidence? Surely confidence reflects the gradual (and hard-earned) accumulation of competence over time? In fact, there is often a mismatch between confidence and competence. In therapy, as in medicine and driving and knowledge of firearms (Dunning et al., 2003), people exhibit a tendency to overestimate their performance. Donald Trump is a very confident man. The most confident.

Therapists, similarly to other humans, exhibit a strong bias towards overrating their effectiveness. Walfish and colleagues (2012) asked a sample of American therapists to rate their performance relative to other clinicians. Every single therapist rated themselves above average. On average, the therapists rated themselves in the top 20%. Only 8.4% of the sample rated themselves below the top 25%.

A similar study with a British sample found similar, albeit more moderate distortions, with therapists rating themselves, on average, in the top 35% (Parker et al., 2015). In a study that compared therapist self-ratings of a session recording with the ratings of an independent rater, Brosan and colleagues (2008) found that the the less competent therapists were more prone to overestimate their abilities.

Given this tendency of humans, and of therapists, to overestimate our competence, it is important to get independent feedback on one’s performance. However, this is exactly what therapists stop doing post-qualification. For example, 80.5% of a sample (n=72) of UK clinical and counselling psychologists said they rely exclusively on self-report during supervision (Nicholas & Goodyear, 2021), meaning no video or audio recordings played, meaning no direct feedback on their actual performance as therapists.

This is a huge missed opportunity, because direct feedback on one’s actual performance offers a possible pathway to improvement. Worse still, the absence of direct feedback may create the conditions in which performance can deteriorate over time. Given the pervasiveness of overconfidence and overoptimism biases, this is likely to occur beyond the limits of one’s awareness.

This overconfidence matters. Constantino and colleagues (2023), found that overestimation of one’s competence is associated with worse outcomes for one’s clients. Conversely, a more accurate self-evaluation, or an underestimation of one’s competence, was associated with better outcomes. Some research suggests professional self-doubt may be conducive to better patient outcomes in some domains (e.g. Nissen-Lie et al., 2013), although findings have been mixed (e.g. Jacobsen et al., 2026).

The longer therapists have been qualified, the better they rate themselves relative to other therapists (Parker et al., 2015). This would make sense if therapist outcomes improved with experience. But they don’t. As such, it could be hypothesised from this that therapists’ increasingly overrate themselves relative to others as they gain more experience.

Very few longitudinal studies have looked at the relationship between experience and self-evaluation. Gonsalvez and colleagues (2023) examined the difference between trainee clinical psychologists’ self-evaluation of their performance, and their supervisors’ evaluations at three points in the training journey. Initially, trainees substantially underestimated their performance relative to their supervisors’ evaluations. However, this discrepancy diminished at the second stage and reversed into an overestimation at the final stage, just before graduation. Would this shift towards overestimating one’s clinical skills continue to become more pronounced post-qualification? Given studies that have recruited qualified therapists show substantial overestimation of skills, this seems possible.

The nature of therapists and drifty drift

One possible link between experience, confidence and gradually deteriorating outcomes is “therapist drift”.

This is the tendency to drift away from the evidence based practices one was trained in. It is a phenomenon that has been noted across multiple contexts and modalities (e.g. Borntrager et al., 2013; Speers et al., 2022). There has been debate about how to define drift. One simple definition characterises it as deviation from evidence based protocols. However, research increasingly suggests that personalisation of treatment improves outcomes (e.g. Nye et al., 2023). As such, this narrow definition of drift has been reframed by some authors as a reflection of “good clinical judgment and a commitment to personalized patient care” (Yonatan-Leus & Tishby, 2025). But therapy can be personalised and remain evidence based (e.g. Moskow et al., 2023).

As such, I’m going to refer here to “drifty drift”, which I’ll define as deviation from evidence-based principles, not protocols. Examples of “drifty drift” would be a psychodynamic psychotherapist avoiding the exploration of transference patterns, or a CBT therapist not implementing exposure for anxiety-related problems, due to beliefs that it will be uncomfortable for clients (Scheveneels et al., 2025), or that one can simply rely on intuition to come up with something better (Gaudiano et al., 2011).

There is some, albeit limited, evidence that therapist experience is linked with drift (Speers et al., 2022). For example, the recency of one’s degree was associated with greater use of evidence-based treatment in youth mental healthcare in the US (Cho et al., 2019), and across a range of psychologists in Australia (Hamill & Wiener, 2018). Waller and colleagues (2012) found that older, and more experienced therapists were less adherent to evidence based practice in CBT for eating disorders. A meta-analysis found that older therapists are less likely to utilise exposure therapy (Langthorne et al., 2023). The authors hypothesised that this could be related to older therapists being more confident and more reliant on clinical judgement than on the evidence base.

Several studies have suggested that experienced therapists are slower to learn new approaches. For example, Beidas and colleagues (2014) found that more experienced clinicians showed poorer fidelity to CBT for youth anxiety after training in the approach. Years of experience (and age) have also been associated with requiring extra training to reach necessary standards in motivational interviewing (Martino et al., 2011).

But a link between greater experience and slower learning of new approaches does not necessarily explain why therapists don’t improve with experience. An experienced therapist could, in theory, continue to improve in their original modality, even if it takes them longer to learn new approaches.

Another possibility is that, as time elapses after one’s training, knowledge fades. This is borne out by a study that found experienced therapists (10 years of clinical experience or more) performed significantly worse than trainees on tests of knowledge, case conceptualisation and treatment planning (Vollmer et al., 2013).

Taking this a step further, is it possible that, as one undertakes more CPD, and picks up new models throughout one’s career, this new knowledge inhibits the retrieval of the old knowledge and training. For example, could it be that my head is so full of recent CFT and DBT and EFT and CAT trainings that it does not consistently and comprehensively, across treatment, retrieve and deploy exposure, for example, in its full and glorious simplicity? Maybe I can’t see the wood for all the -Ts. Maybe I am so intent on making my garden look great that I spend all my time and effort accumulating fancy new gnomes at the expense of attending to the plants that made it a garden in the first place.

Conclusion

Improving therapeutic outcomes is hard. Therapists who wish to do so face many challenges. To summarise…

It is difficult to infer outcomes. In the absence of measures, therapists are overoptimistic about change, both prospectively and retrospectively. They tend to be overconfident about their effectiveness, and this overconfidence potentially grows with experience. Overconfidence can lead to drifty drifting away from evidence based principles, towards watered down, less effective interventions. These issues can gradually erode therapist performance over time. But this doesn’t have to be the case.

Although much remains to be understood about why therapists don’t improve with experience, much is already understood about how they can improve with experience. That will be the topic of the next post.

Did I miss anything?? Share your thoughts in the comments!

Therapists, in their developmental trajectories, are, on average, more like countryside cottages. The roses climb the walls and bloom. The fancy garden gnomes proliferate. So quaint. So appealing. So cottagey. Yet all the while the plaster cracks and the window frames warp and the damp creeps in and there is a gradual (so gradual as to be barely perceptible but it’s there nonetheless) decline in its performance of the primary functions of a house. But there are so many gnomes….

Research suggests that therapists, on average, do not improve with experience and time, and those with more experience do not perform better than those with less.

When I first came across this, I was astonished. I clutched my head in my hands and I rubbed my eyes and I asked: how the hell does experience not lead to better outcomes?

I had always assumed the relationship was positive and mostly linear, with returns diminishing only gradually over time. I had some kind of innate faith in the ability of human beings to learn from experience simply by having that experience. Every experienced therapist I met sounded so good, and so much better than me. They had so many gnomes in their garden. I had assumed that if I kept practicing (as in doing therapy), I would inevitably keep improving. But therapeutic experience does not necessarily bring improvement, and in many cases it coincides with deterioration.

These findings are so important, because they lay the foundations for understanding how therapist outcomes can deteriorate with accumulated experience, and what can be done to help experience contribute to improvement instead.

In this post, I’ll lay out the research that examines the relationship between experience and outcomes. In the next one, I’ll outline some of the possible reasons why experience can be associated with a deterioration in outcomes. In the third post, I’ll explore some of the things that might help to translate experience into better outcomes.

Subscribe now

The effect of therapist experience has long been of interest to researchers. In 1955, Myers and Auld conducted a study examining the psychotherapy outcomes of patients in a psychiatric outpatient clinic, who saw either a resident psychiatrist (less experienced) or a staff psychiatrist (more experienced). The study has tremendous limitations, to the extent the results are barely worth interpreting, but they will be interpreted anyway. The limitations in themselves are quite interesting, and underscore just how far psychotherapy research has come.

This was seemingly before the days of outcome measures, or any attempt to measure outcomes with a veneer of objectivity. Instead, outcomes were determined by looking through therapy notes and inferring an outcome, then categorising it as patient quit/ patient unimproved/ patient improved/ therapy continued elsewhere. For example, the following therapy note was categorised as “patient unimproved”: “In view of the very difficult circumstances in which the patient lives, it seemed that psychotherapy would not have much chance.” The next note earned a “patient improved” label: “I feel she can handle her problems with considerably more freedom than when she came. She seems to have benefited from therapy.”

In this sample of 39 case outcomes across 12 therapists, the authors inferred that the more experienced staff psychiatrists had “more successful terminations and fewer failures than the residents”, suggesting more experienced clinicians attained better outcomes.

However, the authors note, “the differences in the manner of termination between the senior staff and residents may be due to different conceptions of improvement based on differential experience and training”. So there is no sense of a common definition of improvement. Perhaps the more experienced therapists defined improvement more liberally.

The authors offer this beautifully candid caveat:

“We did not make any check on the reliability of our classifications, since it would give a spurious impression of objectivity to study reliability of classification when we have no way of testing the reliability of the raw record.”

I glimpse a flash of red in my garden. I classify it as an orchid, but I don’t look again to double check, because there is absolutely no way of testing whether I might have actually seen a poppy or a dahlia or a ready salted crisp packet pulled from a bin bag by a fox.

There was also something strange going on with case allocation in this service. It seems to be the polar opposite of every service I have ever worked in. In my experience, more senior clinicians take on the seemingly more “challenging” cases, and trainees are allocated seemingly less “complex” cases.

But in this study, the authors write:

“There may also be differences between senior staff and residents in the selection of cases. For instance, it is possible that the senior staff on the basis of clinical experience select cases more likely to be successful and leave the “poor” cases for the residents”.

What a place to train.

As the topic garnered more attention, and was approached with methodologies that offer some impression of objectivity, spurious or not, findings began to coalesce around an absence of improvement with experience.

Minami and colleagues (2009) examined eight years worth of outcome data (measured on the Outcome Questionnaire-45; OQ-45) from a university counselling service. They included data from 2,672 service users seen by 148 different therapists of various orientations, including psychodynamic, cognitive behavioural, interpersonal, humanistic, existential, feminist, and multicultural. These therapists spanned a range of training and qualification levels, which the researchers grouped into “staff” (clinicians who have completed graduate training), “interns” (predoctoral psychology interns or masters of social work/ counselling interns) and “other trainees” (doctoral practicum students on their first clinical placement). The most qualified clinicians (staff) showed inferior effect sizes (d = 0.131 per session1) compared with the interns (d = 0.208 per session) and other trainees (d = 0.218 per session).

However, assignment was not random. The most complex cases were assigned to staff clinicians, so it is reasonable to hypothesise that at least some of the disparity is accounted for by this imbalance.

However, a similar study by Budge and colleagues (2013) with 1318 patients seen by 64 therapists (modality not specified) found similar results even after controlling for clients’ pre-therapy functioning. The interns/post docs achieved more significant change in terms of life functioning and symptom reduction compared with qualified psychologists, measured with the Behavioral Health Measure. In terms of the well-being subscale, there was no significant difference. Trainees achieved similar, or better outcomes, than qualified psychologists.

A smaller study (282 patients seen by 34 therapists) conducted in a specialist, CBT-focused anxiety disorders service found no overall difference in outcomes between trainees and qualified therapists (Mason et al., 2015), although there was an interaction such that qualified therapists attained significantly better outcomes with more severely anxious patients. Notably, the trainees utilised 2.3 fewer sessions than qualified therapists (10.3 vs 12.6). Both Mason and Budge suggested that the level of supervision received by trainees may have contributed to their favourable outcomes. Mason and colleagues (2015) also highlight the freshness of trainees’ theoretical knowledge.

When I was in training, I used to dread asking service users if they would be willing to work with a trainee. I felt like I was asking them if they would accept an inferior service.

But I was always pleasantly surprised (and relieved) by how many people agreed to see a trainee. I don’t recall anyone declining. I wondered whether I would have been so willing to work with a trainee myself. But it seems the service users were wiser than I was. These studies suggest that it would sometimes perhaps make sense to request a trainee instead of a qualified clinician.

Another study found that greater therapist experience (measured as years working in a specialist mental health service) was associated with reduced recovery odds, and less symptom improvement in anxious youth (Fauskanger Bjaastad et al., 2018).

However, a limitation of the studies described so far is their cross-sectional nature. The therapists grouped together could differ on other variables. They could have substantially different levels of prior experience, for example. Later studies have rectified these limitations with longitudinal designs, in which changes in outcomes can be examined within individual therapists as their experience increases, or as they move through successive stages of training.

Goldberg, Rousmaniere and colleagues (2016) analysed data from 6591 service users seen by 170 therapists (mostly integrated/ eclectic). On average, a therapist’s data spanned 4.73 years. They found that therapist outcomes (on the OQ-45) deteriorated as experience (measured as chronological time or as number of cases) increased. The effects remained even when controlling for a range of therapist, patient and caseload variables. For each additional year of practice, on average, a therapist’s effect size would drop by d = 0.012. That is a very small effect, and would amount to, each year, one fewer successful outcome out of 148 patients seen. Yet, over a large population of therapists, that starts to rack up. There was notable variation between therapists. 39.41% of therapists’ outcomes improved over time, while 60.59% deteriorated. This heterogeneity is substantial at the extremes. Some therapists showed dramatic improvement, and some showed mindboggling deterioration. Understanding what causes these differences is crucial, and will be the topic of the next two posts.

Germer and colleagues (2022) replicated this study in the German healthcare system, and found no significant relationship between therapist experience and therapy outcomes.

Erekson and colleagues (2017) conducted a study with similar longitudinal, within-therapist analyses, but focused on change in outcomes as therapists progressed through different stages of training. They found that the amount, and the rate of change therapists were able to achieve either stayed the same (when both stage of training and cumulative number of cases were included in the model), or deteriorated (when only one of the variables was included).

The authors wrote: “Findings suggest that as therapists progress through formal stages of training, they do not improve in their ability to effect change in their clients.”

They add:

“Although this and other studies have implications for experience in general, our study provides persuasive evidence that completion of training over time does not imply improved client outcomes for individual therapists. This indicates a potentially radical implication of these study results—that existing structures for training psychologists, including formal graduate training, internship, postdoctoral appointments, and licensure, do not appear to be connected to improvement in client outcomes or therapy efficiency.”

In another longitudinal study, Coleman and colleagues (2024) looked at outcomes for 42,690 service users seen by 613 therapists across 12 years. The study broke outcomes down into four domains: general distress, life functioning, anxiety and alcohol/ drug use. Therapist outcomes deteriorated across time in terms of general distress, life functioning and anxiety. There was no significant change in terms of alcohol/ drug use outcomes.

Taken together, these studies present a rather curious picture. Therapists, seemingly, do not improve with experience, and do not even improve with the completion of their training. On average, at a certain point in their training, therapists hit a peak which they don’t progress beyond.

However, there are two notable exceptions in the literature (this does not include Myers and Auld and their honorable decision not to “give a spurious impression of objectivity”). Two merciful exceptions that offer an escape from the kind of bleak picture built up so far. These studies created very specific service contexts geared towards therapist improvement. I’ll talk about them more in part three. But I’ll tell you now–there are no gnomes involved.