Quality of Working Alliance in Psychotherapy

Abstract

Therapist characteristics were explored as possible predictors of working alliance, rated early and later in therapy both by therapists (n=59) and patients (n=270) in an ongoing multisite project on process and outcome of psychotherapy. Patients and therapists had divergent perspectives on the working alliance. Therapists' experience, training, skill, and progress as therapists did not have any significant impact on alliance as rated by patients. Training and skill were positively related to alliance as rated by therapists. Interpersonal relationships on the cold–warm dimension had a moderate impact for both patients' and therapists' alliance ratings. Some implications for therapist training are discussed.

In the present study, therapists' professional variables and personal characteristics, based on self-evaluations, were explored as possible pre-treatment predictors of the quality of working alliance in psychotherapy. Working alliance has been a consistent, though modest, predictor of outcome in psychotherapy. In two meta-analyses, the effect sizes (r) have been found to be 0.26¹ and 0.22,² respectively. Early alliance has been found to be a better predictor of outcome than alliance averaged across sessions or measured in the middle or late phase of treatment.³ It should be noted that patients' ratings of working alliance tend to be more highly correlated with outcome of therapy than do therapists' ratings.^1,4 Very few studies have focused on therapists' professional and personal background as predictors. This study is part of the naturalistic Norwegian Multisite Project on Process and Outcome of Psychotherapy (NMSPOP).⁵

The influence of level of experience as a psychotherapist on the therapeutic alliance has been addressed in a limited number of previous studies. In a review of 12 studies,⁶ a small positive relationship between the therapists' experience and the quality of the therapeutic relationship early in treatment was found. In a more recent study, however, therapists' level of experience was not found to be predictive of patients' alliance ratings.⁷ Premature dropout of patients from therapy has been found more frequently with inexperienced than experienced therapists, indicating that experience makes a difference for the therapeutic relationship.⁸ Experienced therapists with training in manualized psychotherapy are likely to contribute positively to the quality of alliance.^9,10 Orlinsky and Howard¹¹ found that therapists who had more than 6 years of experience had a disproportionately high percentage of patients who improved and a disproportionately low percentage of patients who deteriorated during open-ended, relationship-oriented psychotherapy, supporting the importance of the therapeutic relationship. Psychotherapy trainees have been found to develop professional competence in the following order, indicating that working alliance is essential: 1) skills for building rapport with the patients, 2) skills at specific therapeutic techniques and interventions, and 3) ability to articulate a personal theory of psychotherapy.^12,13

In general, previous studies have yielded somewhat variable results with regard to the relationships between therapists' professional experience, formal training, competence as a psychotherapist, and working alliance. The lack of distinction between experience, training, and acquired competence, as well as some variations in ratings methods, may have contributed to reduced consistency in the results, and this also makes comparisons with previous studies difficult.^8,14

The following personal characteristics of therapists have been associated with better working alliance: less self-directed hostility, more perceived social support, and higher degree of comfort with closeness in interpersonal relationships.⁷ Patients of therapists who reported hostile introjects have been found to report no change or negative outcomes in short-term psychotherapy.^15–17 The relationship between complementary patient–therapist behavior and working alliance was explored in a study based on the traditional circumplex model of interpersonal relationships, i.e., with a warm–cold and a dominant–submissive dimension.¹⁸ A positive relationship was found between complementarity as measured from the cold hemisphere: “the closer the dyad fit to perfect complementarity, the stronger the therapeutic alliance as perceived by the patient” (p. 188).¹⁸ In a review of studies of interpersonal complementarity,¹⁹ a repeated finding, however, was that hostile-dominant acts are frequently responded to with further hostile-dominant behavior (i.e., noncomplementary behavior). On the other hand, empathy, nonpossessive warmth, and genuineness are quite likely necessary, often even sufficient, in establishing an optimal therapeutic contact in psychotherapy.²⁰ In a study including the therapists' as well as the patients' developmental history, an association was found between their early parental relations and the therapists' view, but not the patients' view, of the interpersonal process in therapy.²¹

The relationship between therapists' values and the development of working alliance in psychotherapy has not been studied, but previous studies indicate that therapists communicate their values to patients. Therapists' judgments of patients' improvement in psychotherapy have been found to correlate with the extent to which patients appear to adopt the values of their therapists.^22,23 This “convergence of values” suggests that treatment is experienced as effective when the therapy participants begin with somewhat differing value perspectives but close the gap as therapy progresses. In contrast, attitude similarity between therapists and patients has not been found to be related to improvement.²⁴ In a study on value similarity, patients rated sessions as more negative and less engaging if their therapists held dissimilar values.²⁵ Patients' and therapists' perceived similarity of personal characteristics were not correlated with their assessments of progress in therapy.²⁶

The research questions we will explore comprise the relationships between the following therapist variables and working alliance early and later in therapy: 1) professional variables: formal training in psychotherapy, experience, self-rated skill, and progress as a psychotherapist; these variables have not been clearly distinguished in previous studies, but will be analyzed as four separate independent variables in the present study; 2) therapists' self-reports on personal characteristics: interpersonal problems, introjects, and early parental bonding (past relationships); and 3) patient–therapist similarities in personal characteristics and values. No previous studies that are known to us have presented analyses on similarities comprising both patients' and therapists' personal characteristics (such as interpersonal problems, introjects, parental bonding memories, and values) as predictors of working alliance. Previous studies on value convergence give some indication that value similarity may be predictive of working alliance.

HYPOTHESES

1. Because of the inconsistent previous findings regarding the relationships between experience, training, and working alliance, we do not expect to find a consistent pattern of results for these variables and have no specific predictions.

2. Therapists' self-evaluation of progress and skill is related to working alliance: better self-evaluation should be related to higher quality of alliance.

3. A warm interpersonal style in the therapists is related to better quality of working alliance.

4. Therapists' pattern coefficients of introjects (self-attack and self-control) are related to less favorable working alliance.

5. Therapists' memories of warm parental bonding are related to better quality of working alliance.

6. Patient/therapist value similarity may be related to working alliance, whereas the effect of similarity regarding personal characteristics is explored without specific hypotheses.

METHODS

Sample

The Norwegian Multisite Project on Process and Outcome of Psychotherapy (NMSPOP)⁵ is a naturalistic multisite study, with an unselected sample of outpatients (N=270) from seven sites comprising a total of 15 outpatient psychiatric clinics within the Norwegian Public Health system. Six sites offered open-ended individual therapy, whereas one had a limit of 40 sessions. Enrollment of patients into the present study took place from 1995 to 1999. The inclusion policy was liberal: the only exclusion criteria were age below 20, psychosis, drug abuse as the primary problem, mental retardation, and need for emergency treatment and hospitalization. The therapies comprised treatment as usual, mainly within the psychodynamic tradition. The treatments were not manualized. Most therapists had some clinical experience with psychodynamic therapy, as well as postgraduate professional training. By profession, they were 39 clinical psychologists, 13 psychiatrists, 4 social workers, and 3 nurses.

Measurements

In the present study, working alliance was assessed by the Working Alliance Inventory (WAI).^1,27 This instrument was chosen because it is the most widely used in research on alliance. WAI comprises three aspects of working alliance measured in subscales: Bond, Task and Goal, which correspond well to Gaston's definition. Gaston²⁸ attempted to reconcile various definitions of the working alliance, proposing that it is a multidimensional construct composed of four relatively independent dimensions: 1) the patient's capacity to work purposefully in therapy, 2) the patient's affective bond to the therapist, 3) the therapist's empathic understanding and involvement, and 4) the patient–therapist agreement on the treatment goals and tasks.

Therapist Characteristics

Professional Variables. These were assessed by the “Development of Psychotherapists' Common Core Questionnaire” (CCQ).²⁹ The variables were Experience (years of postgraduate clinical practice as a psychotherapist); Train (years of postgraduate formal training in psychotherapy, comprising supervision, personal therapy, and attending a program at an institute for psychotherapy training); Progr (progress as a psychotherapist, comprising therapists' experience of transcendence of past limitations, progress as a therapist, and overall change as a therapist, rated on Likert scales from 0 to 5); and Skill (comprising skills in engaging patients in the working alliance and skills in general theoretical understanding, rated on Likert scales from 0 to 5).

Personal Variables. These comprised IIP-C,^30–32 SASB,³³ PBI,³⁴ and Value Survey,³⁵ and were self-reports, parallel to the versions patients filled out. Current interpersonal relationships were assessed with the Inventory of Interpersonal Problems, IIP-C (64 items, a subset of the 127-item Inventory of Interpersonal Problems). IIP comprises two types of items. The first 39 items begin with the phrase “It is hard for me to…” and the remaining 25 items describe the “things that you do too much.” Each item is rated on a 5-point Likert scale ranging from 0 (not at all) to 4 (very much). Previous research has demonstrated clear circumplex properties of the IIP-C with a thematically meaningful set of eight scales, each with eight items, around the two main dimensions: dominant/submissive and cold/ overly nurturant. Items in the IIP cold subscale are, for example, problems with being too aggressive or too often wishing for revenge. Examples of IIP avoidant (IIP avoid) items are problems with participating in groups, or introducing oneself to other persons. An example of an IIP dominant (IIP dom) item is trying to control other people too much.

Introjects. The Structural Analysis of Social Behavior, SASB (SASB Intrex long form A), Introjects, comprises 36 items describing attitudes and feelings toward oneself, rated on 10-point Likert scales. Two pattern coefficients were computed by correlating the profile of the respondent's cluster scores with ideal or theoretical cluster profiles representing self-attack (SASB att) and self-control (SASB contr). The two pattern coefficients represent degree of similarity in profiles rather than the elevation or mean level of the curve. SASB att measures rejection of self, and SASB contr measures oppression of self.

Past Relationships. The Parental Bonding Instrument (PBI) comprises 25 items for each parent, describing their attitudes and behaviors, rating how one remembers one's parents during the first 16 years of life. Separate ratings are made for mother and father on 4-point Likert scales. The subscales Mother care and Father care measure one of the critical dimensions of parental behavior: expression of warmth versus coldness. The subscales Mother contr (control) and Father contr measure overprotection and control versus allowance of autonomy and independence.

Values. Value were assessed with the Value Survey.³⁵ A value is a conception of something that is personally or socially preferable, and Rokeach³⁵ refers to two kinds of values: Instrumental (I) and terminal (T, end-states). Instrumental values comprise moral values and competence values. They are not necessarily consciously perceived. Examples of instrumental values are tolerance, honesty, obedience, and politeness. Terminal values comprise personal and social values, self-centered or society-centered, intrapersonal or interpersonal in focus; for example, a pleasant life, salvation, national safety, family safety, genuine friendships. The Value survey comprises two subscales, representing the categories of values. Rokeach assumes that the two value scales represent functionally interconnected systems, wherein all of the values concerning modes of behavior are instrumental to the attainment of all the end-state values. The task is to rank the values in the order one prefers.

Procedure

After receiving information and signing a written consent form, the patients had a pre-treatment assessment in two steps: 1) a series of questionnaires was administered, including Symptom Checklist 90–Revised (SCL-90-R),^36,37 IIP-C,^30–32 SASB,³³ PBI,³⁴ Value Survey,³⁵ and Target Complaint (TC)³⁸; and 2) a trained independent clinician did a diagnostic assessment at each site, comprising the DSM-IV SCID I and SCID II (Axis I, II),³⁹ the Global Assessment Scale (GAS),⁴⁰ and a dynamic clinical interview lasting about 1 hour.⁴¹

After the assessment, the patients were scheduled for the first therapy session. According to the design, therapist and patient had not met before. All assessment interviews and therapy sessions were audiotaped. The descriptive data of patients and therapists are presented in Table 1. A score for the total number of personality disorders criteria (SumPD crit) was used, according to the view that cumulative scores of criteria for personality disorders represent the data better than categorical scores.^42–44 The data indicate that the sample of patients was moderately to severely disturbed.^45,46

TABLE 1. Characteristics of patients and therapists.

It is generally agreed upon that the working alliance is sufficiently established to be reliably rated after the third session, and this time was chosen for the first ratings. WAI was subsequently rated after sessions 12 and 20, and then after every 20 sessions throughout the therapy, by both patients and therapists. In the present study we have used data from sessions 3 and 12, from both patients (WAI P) and therapists (WAI T). Total scores of WAI were used. The available ratings were: WAI P3, n=238; WAI P12, n=204; WAI T3, n=132; and WAI T12, n=126. The lower number for the therapists is due to the fact that they started rating WAI one year into the study.

Statistical Analyses

A factor analysis with principal component extraction and varimax rotation was conducted on the 12 WAI items. The results yielded a large general factor, and one small specific factor, corresponding to three of the Bond items. Task/Goal were not found to be unique factors. This is consistent with Hatcher and Barends⁴⁷ but differs from Tracey and Kokotovic's factor structure.⁴⁸ As the general factor explained 54.5% of the variance, and the specific factor only 9%, it was deemed appropriate to use the total WAI score as dependent variable.

Similarity coefficients between patients' and therapists' self-rating instruments (IIP, SASB, PBI, Value Survey) were computed. We used intraclass correlations (Shrout and Fleiss, ICC 1.1⁴⁹) between the parallel data on each item of the scales. This yielded the following variables: IIP-ICC, PBI-ICC, SASB-ICC, Value I-ICC, and Value T-ICC.

For IIP-C, four quadrant coefficients of the circumplex model were computed, according to the main dimensions. The quadrant coefficients were combined from adjacent, related octants: dominant/intrusive (IIP dom), overly nurturant/exploitable (IIP exploit), avoidant/nonassertive (IIP avoid), and cold/vindictive (IIP cold). The first two quadrants are on the warm dimension, and the last two on the cold dimension. This procedure was chosen because the global score would reflect mainly nonspecific interpersonal distress. Using octant coefficients would cause a problem of colinearity in the multivariate analyses because the correlations between the adjacent octant scales are high.^30–32

Initial analyses of demographic variables, gender, age, age difference, and profession did not indicate that these were predictive of working alliance. In the remaining analyses, these variables were not addressed further.

According to procedures for multivariate analyses,⁵⁰ we started with the univariate analyses, examining the simple relations between each potential predictor variable and the dependent variables. The goal was to identify variables that were potential predictors.⁵¹ We wanted to explore all possible, relevant predictors among the different categories of independent variables. We have therefore reported all predictors with P-values less than 0.25 based on correlations with the dependent variables, according to Hosmer and Lemeshow's recommendation.⁵² The univariate analyses determine whether or not each single variable is a significant predictor of working alliance. Some of the most relevant previous studies have presented only univariate analyses, or both univariate and multivariate analyses,^19,22,26 and in line with this, we are presenting both. Multivariate procedures are extensions of univariate analyses⁵³ and are aimed at testing the robustness or redundancy of the predictors.⁵⁴

A strict application of a Bonferroni adjustment for multiple tests would eliminate the statistical significance of our findings. We chose to advocate a less stringent approach that allows a more even balance between Type I and Type II errors,⁵⁵ based on the a priori theoretical and empirical basis of our predictions involving therapists' professional variables, personal characteristics, and past relationships. The effect sizes of the significant findings are modest. Therefore, caution is warranted in drawing conclusions about clinical significance.

After conducting the univariate analyses, a model for hierarchical multiple regression analyses was built from sets of relevant variables with a logical connection to each other.⁵⁵ The model was built from our hypotheses based on previous studies. The 14 independent variables that were included in the hierarchical multiple regression analyses had obtained P-values of less than 0.25.⁵²

If the setwise F was not significant, no t-tests on the set's independent variables were computed,⁵⁵ to protect against large setwise Type I error rates. All patient and therapist responses were assumed to be independent, ignoring the fact that the same therapist treated more than one patient (mean=4.58).²⁶ A significance level of P<0.05 is reported.

RESULTS

The correlations between patient-rated and therapist-rated alliance were low to moderate: r=0.37 (P<0.01) for WAI 3 and r=0.43 (P<0.01) for WAI 12. This result is in line with previous research, and it confirms that patients and therapists have different evaluations of the working alliance. The correlations between WAI 3 and WAI 12 were r=0.68 (P<0.01) for patients and r=0.67 (P<0.01) for therapists. Patients' rating of the alliance were on the average higher than therapists': scores (mean±SD) were 4.94±1.08 vs. 4.66±0.82 for WAI 3 (t=3.14, df=128, P<0.01), and 5.15±0.96 vs. 4.89±0.79 for WAI 12 (t=2.50, df=118, P<0.05). The intercorrelations between independent variables yielded mainly low correlations; nearly 40% were below r=0.10, and only 13% above r=0.40. The highest intercorrelations were found between IIP quadrants, up to r=0.72.

At a general level, none of the therapists' pre-treatment characteristics was strongly related to the quality of working alliance, either as rated by the patient or the therapist, early or later in therapy.

Patients' Ratings of Working Alliance

Univariate Analyses. Professional variables: Quite unexpectedly, more experience and professional training were associated with less favorable alliance (early, later, respectively; Table 2). Interpersonal problems: Problems on the cold dimension (IIP cold) were related to lower quality of working alliance (later). Introjects were not significantly predictive. Past relationships: Warm parental bonding (Father care, Mother care) was related to better working alliance (later), and the same was found for Value similarity (later).

TABLE 2. Correlations between independent and dependent variables.

Multivariate Analyses. The hierarchical multiple regression model was significant only later in therapy. Professional variables were not predictive of alliance. Interpersonal problems on the cold dimension were related to lower quality of working alliance, whereas a more dominant pattern in the therapist was related to better quality. Concerning Introjects, high self-attack had, unexpectedly, a positive association, and self-control had a negative association to working alliance. Past relationships were not predictive, and Value similarity was related to better quality of working alliance (Table 3).

TABLE 3. Hierarchical multiple regression with patient-rated alliance at session 12 as dependent variable.

Combining the univariate and multivariate results, the consistent findings were that the cold dimension of interpersonal problems was related to lower quality of the working alliance, and Value similarity had a positive impact. Introjects were predictive in the multivariate analyses, but not in the univariate. It is difficult to interpret the finding that SASB att was positively related to working alliance. This may be a chance finding or the result of multiple statistical adjustments among variables. Nearly all therapists rated low on self-attack. Only two had relatively high scores. When the hierarchical multiple regression analyses were made without these two scores, self-attack was no longer significantly related to alliance.

Therapists' Ratings of Working Alliance

Univariate Analyses. Professional variables: Skill was related to more favorable alliance both early and later in therapy. Better alliance was also predicted by progress (early) and training (later). Interpersonal problems on both the cold (later) and the warm (early and later) dimensions, and Introjects (self-attack), were related to less favorable alliance (early and later). Past relationships: Mother care was predictive of better quality of alliance (early). Value similarity was not related to alliance (Table 2).

Multivariate Analyses. The hierarchical multiple regression model was significant both early and later in therapy. Professional variables were predictive of alliance ratings both early and later. Skill was the strongest predictor. More training was associated with better quality of working alliance (later), and experience with lower quality (early). Neither Interpersonal problems nor Introjects was predictive. Past relationships (warm parental bonding) were associated with better working alliance (early). Value similarity was not predictive (Table 4).

TABLE 4. Hierarchical multiple regressions with therapist-rated alliance at sessions 3 and 12 as dependent variables.

Combining the univariate and multivariate results, the consistent findings were that Skill and a warm parental bonding predicted better working alliance. In contrast, Interpersonal problems and Introjects were predictive of alliance only in the univariate analyses.

Combining the univariate results for patients and therapists, the only similar findings across participant perspectives were that Interpersonal problems on the cold dimension were related to lower quality of working alliance later in therapy (3rd hypothesis). Furthermore, therapists' warm parental bonding had a positive impact on both participants' assessment of the therapeutic relationship (5th hypothesis). The relationships between training, experience and WAI were inconsistent (1st hypothesis), and not similarly perceived by patients and therapists. Skill and progress were related to working alliance (2nd hypothesis), but only as rated by therapists. Interpersonal problems on the warm dimension had a negative impact on alliance as rated by therapists, but not patients. Introjects were predictive of lower quality of alliance as rated by therapists (4th hypothesis) but not patients. Higher Value similarity was related to better quality of alliance (6th hypothesis) only as rated by patients. Patients ranked the following values highest: family safety, inner harmony, genuine friendships, self-respect, mature love. Therapists ranked the following values highest: family safety, self-respect, genuine friendships, and mature love. Except for inner harmony, the values were identical and the rankings nearly so.

Combining the multivariate results for patients and therapists, we found that the two perspectives of working alliance shared a very small amount of variance, 14% to 18%. The predictors were almost completely different for patients' and therapists' perspectives. Professional variables and past relationships were predictive of alliance as rated by therapists. Interpersonal problems, Introjects, and Value similarity were predictive of alliance as rated by patients.

Our results indicate that therapists' characteristics have some, but a limited, impact on both their own and the patients' evaluation of working alliance. Therapists' self-rated characteristics were weak predictors of patients' ratings of alliance. A somewhat stronger relationship was found between therapists' self-rated characteristics and the same therapists' own evaluations of the alliance, probably because of the lack of independence between these variables. The total amount of variance in working alliance that was accounted for by therapists' characteristics was 31% (early) and 27% (later) for therapists' ratings, and 20% (later) for patients' ratings.

DISCUSSION

Professional Variables

In previous studies on the working alliance, patients' ratings have been found to predict outcome of therapy better, and are therefore probably more valid predictors, than therapists' ratings.^1,4 Furthermore, patients' ratings of alliance and therapists' self-evaluations are independent measures. It is somewhat surprising that longer experience, more professional training, better professional skills, and more progress as a therapist did not have any significant impact on the working alliance as rated by patients. If these variables had any impact at all, the trend was a negative one.

Several interpretations of this result may be possible. First, a common misperception of psychoanalytic clinical theory leads many in the field to view exploratory psychotherapy as the superior treatment modality; dynamic supportive therapy tends to be seen as an inferior method, suitable only for more severely disturbed patients, and is given relatively little attention in training programs. This tendency may lead to the inappropriate use of psychoanalytic techniques in ordinary psychotherapeutic practice.^56,57 In exploratory therapy a neutral stance is emphasized, and it is usually encouraged in training programs. It is possible that experienced dynamic therapists, in order not to compromise the neutral stance, tend to refrain from support, reassurance, education, and role preparation for patients in the beginning of therapy. Many patients, however, prefer active therapists, and a neutral therapist may be perceived as too passive and too little involved. In one study, therapists with a psychodynamic orientation were less effective the more experienced they were, whereas this was not found among therapists with a cognitive-behavioral orientation.⁵⁸ Second, it is also possible that experienced therapists start challenging resistance and defenses too early.^16,17 Third, in contrast to experienced therapists, those with little practice may be eager to avoid frustrating patients or hurting their feelings. Finally, whether some therapists with long experience may feel less enthusiastic about their patients is open to speculation.

Our findings indicate that, generally, outpatients tend to favor the style of less trained and less experienced therapists. Training in dynamic therapy should be alert to possible negative aspects of current training programs and emphasize a more active and supportive attitude in the beginning of therapy. Useful methods may include teaching the patient about the nature of his or her disorder, role preparation for the therapeutic process, and active involvement of the patient in treatment planning. Psychotherapy research has demonstrated that well planned, structured treatment is generally better than unstructured therapy.^8,59

The unexpected negative relationship between experience and early alliance as rated by therapists may simply reflect that experienced therapists are more reluctant to give high ratings to the quality of working alliance early in therapy. They may believe that alliance-building takes more time. More experienced therapists may have increased preparedness for fluctuations and ruptures in the alliance, even if there are signs of a high quality of early alliance. Therapists with long experience may also be reluctant to give higher ratings until they are convinced that the therapeutic process works according to the tasks and goals of therapy.

We found that amount of professional training as reported by therapists themselves was significantly related to better quality of working alliance as rated by the same therapists. Higher working alliance scores by more trained therapists may simply reflect these therapists' confidence in having acquired the skills that facilitate the therapeutic process. However, patients tended to hold the opposite view. Therapists' self-ratings of higher skill and progress as therapists were positively related to alliance rated by the same therapists. This may again reflect the lack of independence between the variables. From the patients' point of view, there were no such associations.

Personal Variables

The cold dimension of the therapist's interpersonal problems was predictive of less favorable working alliance as rated by patients. This was found as a trend also in therapists' alliance scores. The therapists' memories of a caring mother—the warm dimension—had a positive impact on the evaluation of the quality of working alliance from therapists. This was found as a trend also for patients' ratings. Our findings are in line with previous research: the warm dimension facilitates the therapeutic relationship, and the cold dimension has a negative impact. In a separate study with data from NMSPOP, on patient characteristics and the quality of working alliance, the main finding was that patients' interpersonal relations on the cold–warm dimension were the strongest predictors of patient-rated as well as therapist-rated alliance.⁶⁰ A dominant style in the therapists was positively related to the quality of alliance as rated by patients. This may reflect patients' preference for therapists with a more structured, active involvement in therapy, as opposed to a neutral attitude.

Similarities

It is somewhat surprising that we did not find any associations between similarity of personal characteristics and alliance, whereas similarity of values did influence alliance as rated by patients. Values may not be shared explicitly between therapists and patients but may still have an impact on the working alliance. Patients may perceive some cues of the similarities in values and have a positive evaluation of this that is reflected in higher ratings of working alliance. Still, there may be an optimal match between personal similarities/differences. This remains as an issue for future research.

Limitations of the Study

The sample is heterogeneous, the design naturalistic, and therapies were conducted as usual, without adherence checks for therapist interventions. These features may have attenuated our findings compared with those of some other studies. On the other hand, generalizability is stronger with this large, unselected clinical sample of patients and therapists. Late introduction of WAI for therapists limited the number of available ratings. The analyses in the present paper were also made before WAI 12 scores were obtained for the full sample size. All ratings were based on single sessions, which may be less reliable than scores averaged over several sessions.

CONCLUSION

More experience as a psychotherapist is neither a guarantee of better quality of working alliance nor a consistently negative predictor. Therapists' experience, professional training, skills, and progress did not have any significant impact on patients' ratings of working alliance, but a negative trend was found for experience and training. A dominant style in therapists' self-evaluations had a positive impact on alliance as rated by patients, indicating that patients prefer actively involved therapists. Personal similarity variables did not have any impact, but similarity of values predicted the patients' ratings of alliance.

We found that patients and therapists have different perspectives on the working alliance, with only 14% to 18% shared variance. Therapists' self-evaluations on the cold–warm interpersonal dimension predicted alliance as rated by both patients and therapists. We have reported the same pattern in a study on patient pre-treatment characteristics and alliance. These findings are consistent with the main trends in previous research.

Therapist training in dynamic therapy should be alert to possible negative consequences of training programs emphasizing exploratory techniques over supportive techniques. Continued education in psychotherapy may be advisable for experienced therapists also. In the initial phase of the therapy, more emphasis on role preparation, education, reassurance, and support might have a positive impact on the working alliance.

ACKNOWLEDGEMENTS

The study has used data from the Norwegian Multisite Study of Process and Outcome in Psychotherapy, which was supported by grants from the Norwegian Research Council, Medicine and Health, the Norwegian Council for Mental Health, Health and Rehabilitation, and the Department of Psychology, University of Oslo.