The authors would like to acknowledge the useful, patient and detailed editing of Janet Bavelas on this manuscript.

Introduction

When the question of expertise arises in the field of psychotherapy, many authors question whether it is achievable or not (Hill et al., 2017; Shanteau, 1992; Shanteau & Weiss, 2014; Tracey et al., 2014), and some ask how to characterise it (Levitt & Piazza-Bonin, 2016; Norcross & Karpiak, 2017). Ultimately, we need to know what specific actions are performed by an expert psychotherapist (Hill et al., 2017); the importance of research on expert psychotherapists does not reside solely in establishing the effect that experience has on clinical performance. Experts are also models who teach and supervise new psychotherapists (Hill et al., 2015; Hill & Knox, 2013). Therefore, we need to look more closely at what psychotherapists do.

The available research presents some contradictions. Eells et al. (2005) compared the quality of the case formulations of novices with less than 1,000 hours of supervised practice, experienced therapists with more than 10 years of practice, and experts, who had more than 10 years of experience and were also recognised through their publications, manuals or workshops. The ratings of the formulations by the experts were better than for the novices and the experienced therapists but, surprisingly, the ratings of the novices were better than those of the experienced. In the Witteman et al. (2012) study, the master’s-level student group was better in a diagnostic task than practising counsellors. However, research on the outcomes of actual psychotherapy treatment has found no significant differences due to experience (Okiishi et al., 2003, 2006), with outcomes for experienced therapists even becoming poorer in a longitudinal design (Erekson et al., 2017; Goldberg et al., 2016).

Along with this, the results of the meta-analyses are also inconclusive. Walsh et al. (2018) do indicate that greater experience goes hand in hand with greater effectiveness. However, previous meta-analyses yielded results indicating that greater experience brings about less effectiveness (Hattie et al., 1984; Weisz et al., 1995), or that the two variables are not related (Berman & Norton, 1985).

Something that all these studies have in common is that they have not investigated process variables such as use of language (verbal behaviour, communication skills, helping skills, interpersonal skills). According to research, a high level of interpersonal skill (being capable of making clients feel comfortable and understood, thus facilitating the dialogue) is connected with obtaining good clinical outcomes (Anderson, Crowley et al., 2016; Cuijpers et al., 2012; Heinonen, 2014; Kadur et al., 2020), and can even be seen as an essential part of generating predictive models for success (Anderson, McClintock et al., 2016; Schöttke et al., 2017).

These results are not surprising; after all, psychotherapy is a spoken profession (Isaacson, 2019), a conversation between at least two people. The psychotherapeutic context has, however, one characteristic that differentiates it from everyday face-to-face dialogue: therapists present an intentional use of language. This intentional use refers to the fact that theoretical knowledge causes therapists to modify some basic aspects of language, such as the way in which they ask questions and construct formulations in response to the client’s content, highlighting certain aspects, ignoring others, doing so frequently or barely speaking (Jordan et al., 2013; Korman et al., 2013; Tomori & Bavelas, 2007). For example, Huang and Hong (2015) found through microanalysis that skilled SF trainees (not less than five years of experience) used more complex formulations, open questions and positive utterances than novices (less than three months of experience). MacMartin’s (2010) results indicated that solution-focused trainees tend to restructure their questions, when an optimistic presupposition of improvement has not been accepted, to make them more acceptable. Expert CBT therapists seem to perform more interventions with new information, encouraging comments for the client, explanations regarding how the client’s problem works, and greater authority, making direct requests to the client about the tasks to be performed out of session (Froján-Parga et al., 2011; Vargas-Cruz et al., 2014; Vargas-de la Cruz et al., 2018).

All these results should be taken as preliminary studies since, as the authors point out, sample sizes and the descriptive nature of the research do not allow generalisation. The differences found could be due to the therapist’s experience, but also to differences between clients (their problems, their adherence to treatment, etc.), to the idiosyncrasy of each therapist (he or she could be more authoritative because of a personality trait, not because of experience), or to the precise moment of evaluation. It should also be noted that none of the studies reviewed compared the level of responsiveness (Norcross & Wampold, 2018; Stiles & Horvath, 2017) based on experience, nor did they perform analyses that provide insight into the therapist-client interaction.

So far, the research could give the impression that language is an isolated characteristic of psychotherapists. Even the language-coding instruments developed tend to place more emphasis on the role of one (the therapist) of the (at least) two participants (Gumz et al., 2015). When research focuses on the therapist-client interaction, there is some evidence that the psychotherapist and client produce the dialogue together, such that it becomes impossible to understand what one does without taking into account the other, and vice versa (Peräkylä, 2011; Ruiz-Sancho et al., 2013; Stiles & Shapiro, 1995). However, at present we do not have enough evidence about the differences between experts and novices when it comes to client interaction and use of language.

We conducted an exploratory study on these interactions and whether they differ for trainees (or novices) and the supervisor (or expert). Since no consensus has been found on the most accurate definition of experience, expertise or expert (Hill et al., 2017; Norcross & Karpiak, 2017; Shanteau, 1992; Shanteau & Weiss, 2014; Tracey et al., 2014), we have chosen a simple definition that facilitates grouping the participants through a questionnaire. Thus, in this study an expert is a person with more than ten years of experience working continuously as a psychotherapist, who must also be a trainer or supervisor of new therapists. A trainee is a person who has two or fewer years of experience working continuously as a psychotherapist and who is in training at the time his or her performance is recorded.

The main reason for this study is to rethink the process of training psychotherapists. This research argues that by investigating the dialogue between therapists and clients, we may come to understand the process of therapy (Stiles et al., 1998). From this approach, this work aims to understand the differences in the use of language (process) between novice and expert therapists, to make improvements in training procedures. To our knowledge, this is the first such study conducted within a training programme and with this type of sample.

Method

The therapist-client interactions were examined in individual sessions where a novice therapist began the treatment and an expert therapist continued it. This sample is selected with the intention of minimising the alternative explanation that the language used would depend on the case investigated. In this way, the client acts as a constant to investigate the use of language of the expert and the trainees. The design is observational and inter-subject, carried out in naturalistic settings. Descriptive analyses and exploratory hypotheses are performed as a first step toward the study of differences between experts and novices. All participants had given their permission to be video-recorded for research purposes. The ethical review board of the University of Malaga approved the study and all ethical standards were followed (CEUMA: 14-2016-H).

Participants

There were six sessions, each with three participants: a client, a trainee therapist and the expert. The sessions analysed were of an integrative systemic model (Beyebach, 2009; Beyebach & Rodríguez-Morejón, 1999) that has solution-focused brief therapy as its foundation (SFBT; de Shazer, 1985), combined with techniques from MRI therapy (Fisch et al., 1982) and narrative therapy (White & Epston, 1990). Sessions are divided into two parts: an interview where client and therapist share information, and an intervention where the therapist suggests homework or tasks (see Table 1). The video recordings were selected from 23 sessions recorded at a Spanish private psychology centre between 2012 and 2018. Two selection criteria were used: (a) the expert, who supervised trainees and was viewing the session in real time on a monitor, entered the session to continue the treatment’s first phase; and (b) when this happens, the conversation is entirely between the client and the expert, with no verbal contribution from the trainee.

Therapists. The expert was a male, 52 years old, with over 25 years of experience as a psychotherapist. He is a clinical supervisor endorsed by three Spanish professional psychotherapist associations, and the author or co-author of more than 20 book chapters and articles about SFBT. The trainees were six graduate psychologists with no prior experience in psychotherapy. At the time of data collection, they were in their second year of a three-year training programme. In the first year, novice therapists received theoretical training on an integrative model of systemic therapy (Beyebach, 2009; Beyebach & Rodríguez-Morejón, 1999). They carried out at least 90 hours of specific technique practice (for example, miracle question, ineffective solutions, externalisation) and basic interview skills (summaries, backchannels, open-ended questions, etc.). All practical skills are taught through a model where the expert performs the action for students to imitate, receiving constant feedback from their peers and teachers.

Clients. The sample is made up of four women and two men (their characteristics and presenting issues are shown in Table 1). All clients received free treatment.

Data

All databases and extended tables can be accessed from the Open Science Framework repository (https://osf.io/vstxj/). The continuous recording of language was divided into speech turns. These turns are the unit of analysis (the sample) of this research and are defined as: what the client says until the therapist speaks again, and what the therapist says until the client speaks again. This includes all interjections used to maintain the conversation, such as ‘mmmh’ or ‘alright’. The six sessions produced 3,672 speech turns (Table 1).

Instruments

The SICOLENTE is a reliable and valid observational instrument (Rodríguez-Morejón et al., 2018) that consists of 20 categories in three dimensions: Conversational Act (7 categories), Therapeutic Topic (6 categories) and Content (7 categories). It is used with psychotherapy samples recorded in audio or video, in which only the verbal aspect is coded. Each speech turn receives a unique code of three letters, one for each dimension (Table 2).

Procedure

The coding was carried out by the main author. To work with one coder only, the analyst (AZ) was trained with the SICOLENTE manual. We then decided to check the reliability of his coding against a gold standard: the coding sample performed in the intra-coder concordance test of the SICOLENTE development (Rodríguez-Morejón et al., 2018). This procedure implies, as Bakeman and Quera (2011) suggest, the existence of correct and objective coding. According to these authors, a coder is considered reliable if they obtain a Cohen’s kappa of .95. These data include the total number of categories (20), their equiprobability (it was agreed that all categories could occur with the same probability), and the high accuracy sought in the coder (95%).

After obtaining these results, the search for recorded sessions meeting the inclusion criteria was started. Speech turns were coded using LINCE software (Gabín et al., 2012) configured with the SICOLENTE categories. The three dimensions of the instrument were used to code the data. Since the objective of the study is to investigate the therapist-client interaction, only the results obtained from the Conversational Act dimension are reported. The complete data can be retrieved at https://osf.io/vstxj/.

Results

Two statistical analyses were planned: (a) comparison of proportions through Pearson’s chi-square and two-sample Z-tests; (b) one-lag sequential analysis through transitional probabilities (Bakeman & Quera, 2011; Escudero & Rogers, 2004). Following the guidelines set out by Bakeman and Quera, sequential data were analysed only when the row sum was at least 30 and the adjusted residuals were Z ≤ ± 2.58 (p < .01), so as not to overestimate the adjusted residuals and incur a Type I error. To perform both analyses, participants were grouped according to their experience. It is presumed that any differences that could be generated by clients are controlled, since trainees and the expert interacted with the same client.

The descriptive analyses showed that the three most used categories by the novice and expert therapists are Support, Exploration and New information, which constitute 90.2% of what the expert does and 91.8% of what the trainees do. The least used categories are Comment (2.8% for the expert and 2.5% for the trainees) and Exploration introducing new information (7.1% for the expert and 5.8% for the trainees). Taking this into account, the statistical test shows significant differences between the two groups [χ²(4, N = 2015) = 24.292, p < .001]. This first global percentage is broken down differently for each group: the expert presented 25.5% of New information codes and the trainees 17.4%. The Support code was encoded on 60.5% of occasions for trainees, compared to 50% for the expert (Z = 4.46, p < .01).

To continue the analysis, the two main categories that handle common information with the client (Exploration and Support) and the two that introduce new information by the therapist (New information and Exploration introducing new information) were collapsed. Despite the fact that the expert spent less time in session, his ratio of shared information to new information was 422/212 = 1.99, while that of the trainees was 1012/316 = 3.2. This shows that for approximately every two occasions on which the expert talks with the client using common information, he introduces new information on at least one occasion. Trainee therapists, in contrast, introduce new information approximately every three occasions. This difference is statistically significant [χ²(1, N = 1963) = 20.289, p < .001; Z = 4.49].

Client language was similar in both experience groups, with the Follow code used 99.8% of the time when talking to the expert and 99.1% when talking to trainees. The Reject code (which indicates the client’s disagreement with, or clarification of, what the therapist has said) has an extremely low appearance, accounting for only 0.2% of the conversation with the expert (only one occurrence) and 0.9% of the conversation with trainees (12 rejections distributed among four of the six trainees, with a minimum of 2 and a maximum of 4 rejections per session). Table 3 includes descriptive data and the statistical analyses.

Finally, the differences between the expert and the novices were investigated taking advantage of the three-dimensional structure of the SICOLENTE. To do this, categories were collapsed to generate two new ones. The results by dimension indicate that the Neutral (46% for the expert and 52.8% for the novices) and Unspecific (57.7% for the expert and 58% for the novices) categories are the most used in the Therapeutic Topic and Content dimensions respectively. Since these two categories tend to appear when therapists perform backchannels, the idea was to investigate this element in depth. The triad of Support-Neutral-Unspecific codes (‘of course’, ‘aha’, ‘I understand’, ‘ok’) was called Weak supports, and all other Supports (except for the Support-Rules-any category) were called Strong supports (i.e. validating the problem, client goals or improvement, regardless of whether they are behavioural, cognitive, emotional, relational, etc. For example: ‘you’re sad without reason, crying all day, alright’; ‘I understand, before you could be pessimistic too, but now you feel that you’ve lost something, right?’).

Results indicated a significant difference based on experience [χ²(1, N = 1146) = 23.304, p < .001]. Trainees used more Strong supports than the expert, and the expert more Weak supports than the trainees (Z = 2.28, p < .01) (see Table 3).

Regarding sequential analysis, only the trainee-client sequences were significant [χ²(8, N = 1212) = 73.17, p < .01]. Three relational patterns were found. First, the Support code always activates the Follow code in clients (Z = 3.10, p < .01). The other two patterns indicate that whenever novices made changes of meaning (New information or Exploration introducing new information), this activated the client’s Reject code (New information → Reject: Z = 2.85, p < .01; Exploration introducing new information → Reject: Z = 3.78, p < .01).

As indicated, the sequential analysis of the expert with clients was not significant [χ²(4, N = 489) = 2.40, p = .67]. These results were expected considering that, in the contingency table, the combined frequencies of the Reject code for clients display 0 in four cells. This means that, given the testing requirements, statistical analyses cannot be performed. Despite the lack of statistical significance, describing the result is understood to be relevant: there can be no relational pattern with a code that does not occur, in this case the rejection of the expert therapist (see Table 4).

Discussion

To our knowledge, this exploratory study was the first to compare trainee-client communication and expert-client communication in actual therapy sessions with the same client. In this case, in addition to the treatment (all therapists are systemic), the client is constant for all therapists, and that allows us to study the variables of interest (language and experience).

Results showed that the expert therapist performed more interventions intended to modify the clients’ meaning than the trainees did, although he had less contact time with clients. Trainees spent most of the session performing summaries, exploring questions and using backchannels. Furthermore, when the trainees did intervene to modify meanings (e.g., using reframing, deconstructions or establishing new relationships with prior information), clients were significantly more likely to reject this change, according to the sequential analysis results.

An example can be seen in the following trainee-client interaction. The therapist is asking about client exceptions. Language between brackets indicates that both participants overlap when speaking. (1) T: ‘One more thing Julián, we would like to ask if… it ever happens that you… are there moments in your life when you feel closer to saying “hey, well, I’m close to feeling better, today I’m, like, a little more active, a little more positive”?’ (2) C: ‘No, I really never feel like that [and…] I’m always negative like this…’ (3) T: ‘[Never].’ The novice therapist asks a question with a weak presupposition (using the conditional and asking a closed question) about the existence of moments of improvement. After the client’s blunt response, the therapist did not explore these advances again; he accepts what the client has said through a formulation that presents an exactly preserved word (Korman et al., 2013). MacMartin (2010) found a similar response in the sample of systemic novices she investigated: among the possible linguistic strategies, as in our trainee’s example, some novice therapists simply accepted the client’s disagreement with the presupposition and changed the subject.

On the other hand, the sequential analysis of the expert-client interaction did not present statistical significance. However, as discussed in the Results, these data should be interpreted with caution. The chi-square test is not significant because the low frequency of the Reject code prevents calculation in the expert-client interaction. However, the fact that this code does not appear indicates that practically all the new information suggested by the expert therapist was accepted by clients. Overall, the expert therapist receives fewer rejections than novices although he introduces many more changes of meaning (while the expert obtained only one Reject code from the client, novices accumulated up to 4 rejections in a single session).

We might therefore provide two suitable explanations for our results showing client refusals in the trainee-client interaction: they could be understood as evidence of the difference caused by experience, but also as a sign of a bad session seen by the expert, explaining why he decided to enter. More detailed research is needed to understand how the new information-rejection interactions work, since this process could be helpful in shedding light on research with bad outcomes, but also on how the therapist reacts to these moments (Muntigl & Horvath, 2014). Especially for the model of brief systemic therapy studied, being able to introduce changes with which the client does not disagree seems more positive than counterproductive. As Muntigl and Horvath (2014) indicate, these discrepancies compromise the therapeutic relationship and with it the clinical outcomes; they would demonstrate that a proper collaboration between therapist and client is not being achieved (Bordin, 1979).

Regarding supports, novices use more interventions to create therapeutic relationships and, in addition, the type of validations they perform is qualitatively different. Novice supports are longer, including more information such as behaviours, emotions or thoughts that they perceive in the client; these were known as Strong supports. Meanwhile, the expert seeks to validate with shorter interventions, more in the line of backchannels (yeah, right, uh-huh, okay); these were called Weak supports. This can again be understood as part of a strategy focusing more on creating a therapeutic relationship, which seems characteristic of novices. This is perhaps related to the training they have received and have a hard time giving up in their first sessions. The expert therapist, guided by a strategy that focuses more on change, not only diminishes supports but supports the client more concisely and with the objective of maintaining the conversation. This does not mean that the expert stops making empathic comments or worrying about the therapeutic relationship; it could be that the lower frequency of Strong supports merely indicates that the expert knows how to differentiate when it is crucial to validate certain aspects, and which others can be omitted.

Lastly, another plausible explanation for our results is that the expert is not only being responsive to the client’s needs but also to the previous trainee’s performance. Given this, the expert’s results could be elicited by both client and trainee. For example, differences in language use (strong/weak supports, shared/new information) could be related to this reaction of the expert therapist to the previous dialogue between the trainee and the client.

Limitations

In order to carry out this design, with a constant client, some assumptions had to be accepted, and these also imply limitations. This naturalistic study has implications for the sample and its external validity. We researched one psychotherapy model with just one expert. In addition, the small sample size detracts from the solidity of our conclusions and interpretations. Another limitation is that the sequential log-linear analysis used assumes that what is important in dialogue is the relationship between consecutive turns. In other words, we cannot easily analyse the context or track the cumulative effect of previous turns.

Conclusion

This study in a naturalistic setting makes the client a constant, while therapists change. This allows us to study differences in performance between novice and expert therapists. The results are: (1) novice therapists make a greater effort to take care of the relationship, using more support and exploration manoeuvres; (2) the expert therapist introduces more changes of meaning; (3) the expert therapist has a higher proportion of short supports (backchannels) than novices; (4) when novices introduce changes in meaning, they are more likely to be rejected than the expert.

Furthermore, this kind of analysis allows us to obtain feasible feedback for use in training settings. According to our results, during training an effort must be made to improve interventions directed at modifying clients’ meanings, looking to obtain fewer refusals from them, since these interactions could probably correlate with disaffiliation and, therefore, treatment failure (Caspar et al., 2005; Muntigl & Horvath, 2014). This use of language is simple to teach; novices can be encouraged to introduce more changes and perform shorter validations, to emulate the interaction implemented by the expert. When referring to teaching skills, procedures such as role playing and modelling may be the most appropriate (Hill & Knox, 2013). The fundamental aspect in training is that novices may become capable of better adjusting to clients (Stiles & Horvath, 2017). Along with the use of language, training should seek to make novices feel effective and skilled, since all three aspects predict good clinical outcomes, especially in brief therapists (Heinonen, 2014, p. 64). In addition, it seems sensible to think that they are related: a greater sense of efficiency and skill in therapy, more attempts to change and introduce meanings, and less insecurity about the state of the therapeutic relationship.