Referred to variously as the therapeutic, working, or helping alliance, the idea that the therapeutic relationship is important for therapeutic success was foreshadowed by Freud's (1) comments on the positive feelings that develop between doctor and patient. Subsequent psychoanalytic writers such as Greenson (2) and Zetzel (3) articulated this concept more fully, distinguishing between the "real" and adaptive dimension of the treatment relationship and the transferential and fantasy-laden aspect. In his Client-Centered Therapy, Rogers (4) identified the empathic bond between the patient and therapist as the essential therapeutic agent in treatment.

Seeking to operationalize this concept and apply it more generally across psychotherapies, Bordin (5) identified three components of the therapeutic alliance—goal, task, and bond. He saw the therapeutic alliance as a mutual construction of the patient and therapist that includes shared goals, accepted recognition of the tasks each person is to perform in the relationship, and an attachment bond. He saw the therapeutic alliance as developing in the relationship between the two and as the vehicle through which psychotherapies are effective. He noted that different psychotherapies call upon different aspects of the therapeutic alliance at different points over the course of treatment.

Although the concept of the alliance has emerged historically in the psychodynamic literature, the strength of the collaborative relationship between patient and therapist has been recognized as crucial by therapists from different theoretical backgrounds. Most theorists, including Beck et al. (6), emphasize the establishment of the patient-therapist relationship as an important first step of treatment. There is evidence suggesting that the therapeutic impact of the alliance is similar across diverse forms of treatment (7,8), although it might differ as a result of which alliance measure is used (7).

Measuring Therapeutic Alliance

Luborsky's Penn Helping Alliance scales (9) were the first therapeutic alliance measures developed. The widely used Penn Helping Alliance questionnaire (HAq) was recently revised into a 19-item scale (10) because the earlier questionnaire included some scale items which reflected how much the patient had benefited from therapy. Some researchers suggested that the measurement of the alliance should be as independent as possible of how much the patient had already benefited from treatment. Thus, six out of the 11 original HAq items were eliminated and 14 new items were added in the revised 19-item questionaire in which each item is rated on a 6-point Likert scale.

Horvath and Greenberg's Working Alliance Inventory (11) is a three subscale measure based on Bordin's (5) tripartite definition of the alliance: agreement on tasks and goals and the patient-therapist bond. The self-report and observer-rater versions use 7-point Likert type scales in a 36-item measure and also a shorter 12-item version. The three subscales of 12 items each are highly intercorrelated (11). The observer version of the WAI has been found to have good interrater reliability (12,14) and to correlate highly with other measures of alliance such as the California Psychotherapy Alliance Scales (CALPAS) (15) but to correlate more moderately with the Penn Helping Alliance Scale (12).

The CALPAS (15) is a 24-item self-report questionnaire that measures patient-therapist therapeutic alliance as a multidimensional construct. The CALPAS four subscales assess (16): (i) the patient's capacity to work purposefully in therapy, (ii) the affective bond with the therapist, (iii) the therapist's empathic understanding and involvement, and (iv) the agreement between patient and therapist on the goals and tasks of treatment. Each item is rated on a 6-point Likert scale. Relatively high intercorrelations among CALPAS subscales have also commonly been observed (12,13,17,18), leading many investigators to primarily use the total score. Acceptable interrater reliabilities were obtained for the observer version (12,14), which has been found to be highly intercorrelated with the WAI-O (12). The CALPAS may have adequate predictive validity in different therapies, including cognitive behavioral therapy (14), psychodynamic psychotherapy (19), and across different treatments (20), especially among "neurotic" patients, but it may be a weak predictor of outcome with cocaine-dependent patients (17).

Each of the three major therapeutic alliance measures—the Penn Helping Alliance, the Working Alliance Inventory, and the CALPAS—includes a therapist and patient rated version as well as an independent observer version. There is clear evidence that patient self-report is a better predictor of outcome than therapist self-report (8), especially when assessed early in treatment.

On average, the effect size of the correlation between the therapeutic alliance and therapeutic outcome is 0.22, based on a meta-analysis of 79 studies (7). Some researchers have suggested that the alliance explains most of the systematic variance in outcome and that it explains more of that variance than other variables (21,22).

In summary, there are many scales for measuring therapeutic alliance, and we have presented the basic psychometric data for the three scales most widely used and most likely to be used in training of residents. These scales predict therapeutic outcome, with the patient report version providing the most robust data.

Therapeutic Alliance Skill Development

Patient qualities, therapist qualities, and therapist "technical activity" are the broad categories of factors that are thought to affect the development of the therapeutic alliance. Moras and Strupp (23) noted that 25% of the variance in a patient's collaborative participation in therapy is linked to antecedent patient variables such as the nature and quality of the patient's interpersonal relationships. Satterfield and Lyddon found that prior dependent attachment relationships predict a negative view of the therapeutic relationship (24). Patients' pretreatment expectations of improvement predicted a better therapeutic alliance early in therapy, while hostile, dominant interpersonal problems predicted a poorer alliance (25). Regarding therapist qualities, Dunkle and Friedlander (26) found that less self-directed hostility in the therapist, more perceived social support, and comfort with closeness led to a stronger bond component of the therapeutic alliance.

The therapist "technical activity" represents perhaps the most teachable component of the therapeutic alliance, and therefore the focus of this communication. We review contributions from the general psychotherapy training literature, studies of the role of specific skill acquisition and clinical/training experience in therapeutic alliance-building skills, and developmental studies of trainees' therapeutic alliance skills.

In their early review of the general therapist training literature Albert et al. (27) concluded that little progress had been made in training therapists to attend to the interpersonal processes in the treatment relationship, and that there was little evidence that training-related changes in therapist technical activity were sustained over time. Lambert (28) found that learning the technique of Rogers' Client-Centered Therapy did not improve patient outcome.

Other contributors have emphasized the role of specific skill acquisition and clinical experience in learning how to develop a therapeutic alliance with a patient. Grace et al. (29) demonstrated that counselor trainees who were taught to explicitly discuss patient nonverbal communication had improved therapeutic alliance scores when compared to trainees who simply expressed empathy. Weiden and Havens (30) identified specific behavioral techniques for improving the therapeutic relationship with severely disturbed patients. Crits-Christoph et al. (31) reported that accurately interpreting patients' core conflict early in treatment results in increased therapeutic alliance later on in treatment. Safran (32) noted that the ability to repair the inevitable ruptures in the therapeutic relationship is essential for strengthening the therapeutic relationship.

Therapeutic alliance skills may also develop with clinical experience and duration of training. Mallinckrodt et al. (33) were the first to specifically look at the relationship between training and measurable therapeutic alliance in the Bordin model. Studying three cohorts of counselors, including early trainees, those later in training, and experienced practitioners, Mallinckrodt et al. found that greater experience is associated with higher goal scores, less powerfully with improvement in task scores, and is not correlated with bond scores. Dunkle and Friedlander (26) found that training experiences did not predict increased goal and task scores. Davenport's (34) study of family therapy trainees found that therapeutic alliance was correlated with cumulative number of patient hours and was associated with how closely trainees were meeting the program's clinical hour requirements.

Kurcias' (35) qualitative study of psychology trainees and their supervisors' conceptualization and implementation of the therapeutic alliance suggested an evolution in the trainees' work. Trainees showed increased sophistication, complexity and focus in their conceptualization of the patient and of the alliance, greater comfort in discussing patient-therapist relationship issues, greater patience with the slow pace of change, and increased recognition of and more skillful management of relationship ruptures over the course of their training.

All of the work described above involves psychology or counseling trainees, and it is interesting to consider the potential factors that may affect therapeutic alliance building skill development in psychiatry residents. The early and intensive exposure to the medical doctor-patient relationship, parallel learning about the psychopharmacologist-patient relationship, the experience of learning multiple therapist-patient roles simultaneously, and the potentially greater average age of trainees may affect the development of this skill.

We conclude from the work on therapeutic alliance skill development that: 1) the ability to develop a therapeutic alliance is a skill that can develop during training and may improve through acquisition of specific skills, accumulated clinical hours, and more complex case conceptualization, 2) trainees become more focused on the therapeutic alliance with greater training and/or clinical experience, 3) certain aspects of the alliance such as goal setting and task recognition may be more learnable and teachable than are other aspects such as bond development, and 4) there are preexisting therapist factors that affect the ability to develop a therapeutic alliance.

Evaluating Therapeutic Alliance Pedagogy

Understanding the predictive value of therapeutic alliance and aware of trainee skill development in this area, we will review the small literature on therapeutic alliance pedagogy. Beitman's (36) modular psychotherapy teaching program places therapeutic alliance development in a central role, exposing residents to the concept early in the course of psychotherapy training. He reviews the main theoretical concepts, accepting Bordin's definition, and identifies the goal of increasing trainee self-awareness through observation and systematic rating of the therapeutic alliance. Trainees rate the working alliance they have with two patients and in two standard video interviews. Through these exercises, the trainee becomes more aware of the importance of the therapeutic alliance, and specifically of the role of task, goal, and bond.

Grace (29) offered a 15-week counseling methods course that focused on specific therapy techniques for training counselors to explicitly address nonverbal behavior of their patients. As previously discussed, this resulted in improved therapeutic alliance ratings. Davenport (33) measured the therapeutic alliance of marriage and family therapy trainees and found that those below the program clinical hours requirement tended to have lower alliance scores. She noted that there will be significant variance in scores across an individual therapist's patients and also noted the ability of trainees to find at least one "good patient" who rated the therapeutic alliance highly. Kurcias (35) observed the teaching methods used for her cohort of traineesvand underscored the crucial role of individual supervision on the initial recognition of the importance of the therapeutic alliance, and subsequently on the complex formulation of the patient and the treatment relationship and on handling relationship ruptures. She found that there was relatively little didactic attention to this area. Finally, Binder et al. (37) proposed a series of recommendations for changes in psychotherapy teaching that included increased focus on "basic relationship skills," and research into the teachability of relationship skills.

It is interesting to consider whether any of the identified therapist factors in the therapeutic alliance may be modifiable, that is, whether these contributions to the alliance could benefit from training. Could residents' own psychotherapy experiences improve their ability in this area? Satterfield's (24) finding that self-directed hostility and attachment issues decrease the bond component of the therapeutic alliance suggests that perhaps successful therapy for some psychotherapy trainees offers the possibility of improving their ability to foster the bond dimension of the therapeutic relationship.

Only two training programs have reported using therapeutic alliance data in the training. Beitman (35) uses the data to provide feedback to help trainees improve their observation skills. Davenport (33) uses the data to identify trainees who are having difficulty. There are no programs that utilize therapeutic alliance data to evaluate their pedagogy in this area.

We conclude from this review of therapeutic alliance pedagogy that therapeutic alliance data may be a valuable part of the educational process. The pairing of patient data with therapist data may provide important feedback to the therapist, as well as important data to the faculty of the training program about a residents' skills in this area. It may also be valuable information for use in individual supervision. The more teachable and learnable parts of therapeutic alliance building may be the ability to effectively set shared goals, and identify appropriate tasks in the treatment relationship, while the ability to foster an attachment bond may be more dependent on antecedent therapist variables.

Therapeutic Alliance Measure and Psychotherapy Competency

Because the therapeutic alliance concept and its measurement stands out in the psychotherapy literature as the strongest predictor of psychotherapy outcome, and there is understandable concern over the untested validity and reliability of the many psychotherapy competency procedures being implemented in the wake of the revised Residency Review Committee Requirements (38), we recommend that psychiatry residency programs consider measuring therapeutic alliance as a criterion for competency and as a tool for providing feedback to residents. It is our hope that this limited review of the alliance measures will be helpful to program directors in developing such systems.

The purpose of instituting psychotherapy competencies was to set a bar for the profession, to encourage programs to monitor their own psychotherapy training outcomes, and to evaluate and improve their training effectiveness. We suggest that resident's therapeutic alliance data will be an important window into the psychotherapy training process. We recommend the following pedagogical approaches to improving therapeutic alliance skills: 1) an early didactic and supervisory focus on therapeutic alliance concepts and techniques, 2) sustained attention later in training on formulation and case conceptualization, with supervisory attention to anticipated relationship ruptures, 3) didactic and supervisory attention to the setting of appropriate, realistic, and negotiated goals in therapy and the identification of patient and therapist tasks in therapy, and finally 4) integration of the concept of therapeutic alliance development with data about this into case conceptualization and formulation teaching.

We also make the following related programmatic recommendations: 1) evaluation of psychotherapy training programs in terms of when the goal, task, and bond components of the therapeutic alliance may be addressed educationally and 2) collection of therapeutic alliance data for educational purposes, competency assessment and program evaluation. Based on experience in the clinical research setting, we recommend alliance ratings at sessions 2, 5, and 10, followed by every 3 months ratings for longer psychotherapies. We anticipate that three to five patients' ratings are probably adequate, and if more than one patient reports an inadequate alliance, a larger patient sample may be indicated.

Using therapeutic alliance data to evaluate competency raises several issues. Should a threshold be established for competence, and what degree of variance should be anticipated? Is the data useful for evaluating excellence as well as competence? What are the effects on residents and on patients of systematic collection of therapeutic alliance data? Researchers' experience (17,19) has been that, on average, most patients tend to rate alliance with therapists quite highly; therefore any low rating should be attended to closely. Similarly, when there are big discrepancies between the therapists' and patients' alliance ratings in medium- to long-term therapy, closer examination of the alliance may be warranted.

Research is lacking the developmental and training issues involved in learning therapeutic alliance skills and in the efficacy of therapeutic alliance pedagogy in psychiatry residency programs. The therapeutic alliance in the psychopharmacology treatment relationship is a subject of further study of particular interest in psychiatry residency training.