Therapeutic alliance and outcome of psychotherapy: historical excursus, measurements, and prospects for research.

This paper proposes a historical excursus of studies that have investigated the therapeutic alliance and the relationship between this dimension and outcome in psychotherapy. A summary of how the concept of alliance has evolved over time and the more popular alliance measures used in literature to assess the level of alliance are presented. The proposal of a therapeutic alliance characterized by a variable pattern over the course of treatment is also examined. The emerging picture suggests that the quality of the client-therapist alliance is a reliable predictor of positive clinical outcome independent of the variety of psychotherapy approaches and outcome measures. In our opinion, with regard to the relationship between the therapeutic alliance and outcome of psychotherapy, future research should pay special attention to the comparison between patients' and therapists' assessments of the therapeutic alliance. This topic, along with a detailed examination of the relationship between the psychological disorder being treated and the therapeutic alliance, will be the subject of future research projects.

Introduction

The main aim of this paper is to propose a historical excursus of the most relevant literature which has investigated the relationship between the therapeutic alliance and outcome in psychotherapy.

A challenge by Eysenck (1952), who claimed that the efficacy of psychotherapy had not been demonstrated and that any improvements were the result of so-called spontaneous remission, stimulated significant developments in the study of outcomes in psychotherapy. Furthermore, research into the relationship between the process and outcome of psychotherapy has frequently attempted to explain the non-specific factors theorized by Strupp and Hadley (1979) which can have a significant impact on the outcome of different treatments. This viewpoint was more recently confirmed by Strupp (2001), who showed that the outcome of a psychotherapeutic process is often influenced by so-called non-specific factors, namely, the personal characteristics of the therapist and the positive feelings that arise in the patient – feelings which can lead to the creation of a positive therapeutic climate from an emotional and interpersonal perspective.

From a different perspective, Orlinsky and Howard (1986), in their review of the research into process and outcome in psychotherapy, seek to respond to the following question: what is effectively therapeutic about psychotherapy? Here, it is important to note that research in the field of psychotherapy is usually classified as outcome research and process research. Outcome research analyses the results of the therapy, whereas process research investigates the various aspects of the therapeutic process, which can also be measured during the course of therapy regardless of outcome. This process is what takes place between, and within, the patient and therapist during the course of their interaction (Orlinsky and Howard, 1986). These two areas of research should not really be considered as separate, but rather as two sides of a coin. Migone (1996) distinguishes three partially overlapping phases in the history of psychotherapy research: a first phase, between the 1950s and 1970s, when research focused on the outcome of psychotherapy and there was a proliferation of meta-analysis; a second phase between the 1960s and 1980s in which there was a growing interest for research into the relationship between process and outcome (the Vanderbilt Project is the most famous example of this); and a third phase from the 1970s onward, in which interest shifted to the therapeutic process and the desire for a greater understanding of the “micro-processes” involved in therapy.

Before examining the most influential instruments designed to measure the therapeutic alliance and their correlations with outcome, we will summarize the concept of alliance as it has evolved over time.

Evolution of the Concept of Therapeutic Alliance

According to Horvath and Luborsky (1993), the concept of therapeutic alliance can be traced back to Freud’s (1913) theorization of transference. Initially regarded as purely negative, Freud, in his later works, adopted a different stance on the issue of transference and considered the possibility of a beneficial attachment actually developing between therapist and patient, and not as a projection. Along the same lines, Zetzel (1956) defines the therapeutic alliance as a non-neurotic and non-transferential relational component established between patient and therapist. It allows the patient to follow the therapist and use his or her interpretations. Similarly, Greenson (1965) defines the working alliance as a reality-based collaboration between patient and therapist. Other authors (Horwitz, 1974; Bowlby, 1988), expanding on the concept of Bibring (1937), considered the attachment between therapist and patient as qualitatively different to that based on childhood experiences. These authors made a distinction between transference and the therapeutic (or working) alliance, and this distinction later extended beyond the analytical framework (Horvath and Luborsky, 1993).

Rogers (1951) defines what he considered to be the active components in the therapeutic relationship: empathy, congruence, and unconditional positive regard. These were seen as the ideal conditions offered by the therapist but were later shown to be specifically essential for client-centered therapy (Horvath and Greenberg, 1989; Horvath and Luborsky, 1993). While Rogers stressed the therapist’s role in the relationship, other works focused on the theory of the influence of social aspects. The work of Strong (1968) was based on the hypothesis that if the patient is convinced of the therapist’s competence and adherence, this will give the latter the necessary influence to bring about changes in the patient.

Recognition of the fact that different types of psychotherapy often reveal similar results gave rise to the hypotheses regarding the existence of variables common to all forms of therapy, rekindling interest in the alliance as a non-specific variable. Luborsky (1976) proposes a theoretical development of the concept of alliance, suggesting that the variations in the different phases of therapy could be accounted for by virtue of the dynamic nature of the alliance. He distinguished two types of alliance: the first, found in the early phases of therapy, was based on the patient’s perception of the therapist as supportive, and a second type, more typical of later phases in the therapy, represented the collaborative relationship between patient and therapist to overcome the patient’s problems – a sharing of responsibility in working to achieve the goals of the therapy and a sense of communion.

The definition of the therapeutic alliance proposed by Bordin (1979) is applicable to any therapeutic approach and for this reason is defined by Horvath and Luborsky (1993) as the “pan-theoretical concept.” Bordin’s formulation underlines the collaborative relationship between patient and therapist in the common fight to overcome the patient’s suffering and self-destructive behavior. According to the author, the therapeutic alliance consists of three essential elements: agreement on the goals of the treatment, agreement on the tasks, and the development of a personal bond made up of reciprocal positive feelings. In short, the optimal therapeutic alliance is achieved when patient and therapist share beliefs with regard to the goals of the treatment and view the methods used to achieve these as efficacious and relevant. Both actors accept to undertake and follow through their specific tasks. The other two components of the alliance can only develop if there is a personal relationship of confidence and regard, since any agreement on goals and tasks requires the patient to believe in the therapist’s ability to help him/her and the therapist in turn must be confident in the patient’s resources. Bordin also suggests that the alliance will influence outcome, not because it is healing in its own right, but as an ingredient which enables the patient to accept, follow, and believe in the treatment. This definition offers an alternative to the previous dichotomy between the therapeutic process and intervention procedures, considering them interdependent.

Only a few studies have examined the relationship between alliance and outcome in group psychotherapy. One conceptualization of therapeutic alliance in group psychotherapy follows Bordin’s theory, transferring this multifactorial construct from an individual to a group setting. The first difference is that in group psychotherapy we have multiple therapeutic agents: the therapist (usually two co-therapists), the members of the group, and the group as a whole. Thus, we have to consider more than one relational level within the group: member to therapist alliance (the same as individual therapy), member to member alliance, group to therapist alliance, and member to other members as a whole alliance. Under this complexity of adapting the alliance concept to a group context, some authors have found a solution: the systemic model of alliance according to Pinsof (1988) Pinsof and Catherall (1986). These authors have adapted Bordin’s model to multiple interpersonal subsystems. These subsystems involve (a) a self-to-therapist alliance, (b) group-to-therapist alliance, (c) self-to-members alliance, and (d) other-to-therapist alliance. Under this point of view, an alliance can be conceptualized as the totality of the alliances formed (Gillaspy et al., 2002).

In a comparison of therapeutic factors in group and individual treatment processes by Holmes and Kivlighan (2000), relationship components have emerged as being more prominent in group psychotherapy, whereas emotional awareness–insight and problem definition change are more central to the process of individual treatment. As such, we can say that clients in group therapies may attach greater importance to relationship factors.

When defining therapeutic alliance in a group context, it is necessary to take into account the comparison with group cohesion, another central construct that is often confused with alliance. Definitions of cohesion have covered a wide range of features, sometimes overlapping the alliance construct. Yalom (1995) speaks of a sense of support, trust, belonging in the group, and also “the analog of relationship in individual therapy”; Budman et al. (1989) refer to cohesion as working together toward a therapeutic goal and engagement around common themes. They found that alliance and group cohesion were closely related and that both were strongly related to improved self-esteem and reduced symptomatology. Crowe and Grenyer (2008) make a distinction between cohesion and alliance, stating that group cohesion refers to the relationship between all members of the group, including the therapists (Burlingame et al., 2011), while working alliance, by contrast, refers to the relationship between the therapist and group member. Marziali et al. (1997) tested the contribution of therapeutic alliance and group cohesion (both based on self-report) to outcome in group therapies for borderline personality disorder. Cohesion and alliance were correlated significantly and both predicted a successful outcome, although the alliance accounted for more outcome variance.

Measuring the Alliance

Table 1 shows the alliance measures more frequently used to assess the level of alliance and their correlations with outcome. Most of them are based on the theoretical assumptions previously described.

Table 1

The most common alliance measures available in literature for adult psychotherapy.

Measure	Therapeutic model and description	Scoring system	Therapists’, clients’, and/or clinical observers’ perspective	Psychometric properties	Correlations with outcome
Pennsylvania (Penn) scales(Alexander and Luborsky, 1986; Luborsky et al., 1985; Luborsky et al., 1983).	Luborsky’s (1976) psychodynamic concept of the two types of helping alliance (i.e., patient’s perception of the therapist as supportive, and representing the collaborative relationship between patient and therapist to overcome the patient’s problems). These scales rate the alliance and consist of the HAcs (Helping Alliance Counting Signs Method), HAr (Helping Alliance Rating Method), and HAq (Helping Alliance Questionnaire Method).	A score on a 5- (HAcs), a 10- (HAr), and a 6- (HAq) point rating scale is assigned to a series of items grouped into sub-scales according to the type of alliance being considered. For each scale, the alliance score is the sum of the subscale ratings.	Therapists (HAr)/clients (HAq)/clinical observers (HAcs)	High level of internal consistency (0.93) and less robust convergent validity compared with other measurements of alliance (Elvins and Green, 2008).	These scales have been shown to be moderately correlate with outcome (r = 0.29; Luborsky et al., 1983; Martin, Garske and Davis, 2000]) also independently of pre-treatment characteristics such as severity of illness (Le Bloch et al., 2006).
Vanderbilt scales(Hartley and Strupp, 1983; O’Malley et al., 1983; Suh et al., 1986)	Strupp’s dynamic and integrative conceptualizations of alliance (Strupp and Binder, 1984). The VPPS (Vanderbilt Psychotherapy Process Scale) measures the relationship between therapist and patient and the psychotherapy process. This scale was later modified to become the VTAS (Vanderbilt Therapeutic Alliance Scale) more specifically designed to measure the therapeutic alliance.	In the VPPS rating is performed on a segment of the therapy, using a five-point scale to measure 80 items. In the VTAS tapes of treatment sessions are rated using a six-point scale to measure 44 items.	Clinical observers	A factor analysis conducted on the two scales found that the VPPS and the VTAS had similar factor structures (Hartley and Strupp, 1983). The VTAS has demonstrated solid inter-rater reliability, internal consistency, and convergent validity with other alliance measures (Krupnick et al., 1996).	These scales have been shown to be moderately correlate with outcome (r = 0.25; Martin et al., 2000).
Toronto scales(Marziali, 1984; Marziali et al., 1981).	Classic psychodynamic conceptualizations of the alliance as well as Bordin’s (1979 integrative model. Specific focus on the affective aspects of the alliance. By combining items taken from other scales (VPPS, VTAS, and HAcs) Marziali and colleagues developed the TARS (Therapeutic Alliance Rating Scale). There are three versions of the TARS according to the rater’s perspective	All of the three versions of the TARS consist of 42 items (21 pertaining to the patient and 21 pertaining to the therapist). Each item is rated on a six-point scale.	Therapists/clients/clinical observers	Adequate internal consistencies and convergent validity with self report versions of the Penn scales and VPPS (Elvins and Green, 2008).	These scales do not appear to be related to outcome (r = 0.07; Martin et al., 2000).
Working alliance inventory (WAI)Horvath and Greenberg 1986) (1986, 1989)	The WAI measures the quality of the alliance on the basis of the three aspects of the alliance theorized by Bordin’s (1979): the bond, the agreement on goals, and the agreement on tasks. There are three versions of the WAI according to the rater’s perspective. Tracey and Kokotovic (1989) have developed a shortened version of these scales.	The WAI is a self-report scale consisting of 36 item each of one rated on a seven-point scale. The shorter version consists of 12 item.	Therapists/clients/clinical observers	Strong support for the reliability of the WAI scales and some support for its validity. Several studies have demonstrated the predictive validity of this instrument in a variety of treatments (Horvath, 1994; Horvath and Greenberg, 1989; Howard et al., 2006; Klein et al., 2003; Martin et al., 2000; Safran and Wallner, 1991).	These scales have been shown to be moderately correlate with outcome (r =0.24; Martin et al., 2000).
California scales(Gaston and Marmar, 1994; Marmar et al., 1989a; Marmar et al., 1989b)	The California Scales comprise the CALTARS (California Therapeutic Alliance Rating Scale) and the CALPAS (California Psychotherapy Alliance Scale). The former derives from the TARS and focuses on the affective and attitudinal aspects of the alliance rather than on specific therapeutic interventions. The CALPAS is a revised version of the CALTARS and was designed to rate the four aspects of the alliance identified by Gaston (1990): patient working capacity, patient commitment, therapist understanding and involvement, patient–therapist agreement on goals, and strategies.	The CALTRAS consists of 41 items, 20 of which refer to the therapist, and 21 to the patient. The CALPAS is a self-report 24-item questionnaire. Each item is rated on a seven-point scale.	Therapists/clients/clinical observers	Factor analytic studies have shown confirmation for the four aspects of the alliance identified (Marmar et al., 1989a,b). The CALPAS is highly correlated with the WAI.	These scales have been shown to be moderately correlate with outcome (r = 0.17; Martin et al., 2000; Safran and Wallner, 1991).
Therapeutic session report (TSR)Orlinsky and Howard (1966, 1986)	The TSR measures the three dimensions of the therapeutic bond as defined by Orlinsky and Howard (1986): working alliance (investment of patient and therapist into their respective roles), empathic resonance (the participants’ joint sense of being understood by each other) and mutual affirmation (the experience of a reciprocal, caring attitude). In developing this instrument, Orlinsky and Howard were influenced by the Rogers’s (1951) notion of unconditional positive regard.	The TSR is a 145-item structured-response instrument. Most of the item are scored in a binary fashion or on a 0–2 scale.	Therapists/clients experience is a valid indicator of alliance quality (Saunders, 1999).	Adequate internal consistency and inter-rater reliability (Elvins and Green, 2008; Kolden, 1991). The results show that the patient’s in-session emotional	No correlations between alliance and outcome (Elvins and Green, 2008).
Therapeutic bond scales (TBS)(Saunders et al., 1989)	Based on Orlinsky and Howard’s (1986) psychotherapeutic model. The TBS are developed from the theoretical assumptions of the TSR.	This instrument consists of 50 item belonging to the following dimensions: 15 items compose the Working Alliance scale, 17 items compose the Empathic Resonance scale, and 18 items compose the Mutual Affirmation scale. Altogether, these subscales provide a Global Bond scale. Each item is rated on a 21-point scale.	Clients/clinical observers	The internal reliabilities of each subscale is adequate, as is the internal reliability of the Global Bond scale. All three scales and the Global Bond scale are related to patient ratings of session quality (Martin et al., 2000; Saunders et al., 1989).	The Global Bond scale is related to outcome (r = 0.37) but there are limited correlations studies (Martin et al., 2000).
Psychotherapy status report (PSR)(Frank and Gunderson, 1990)	The PSR was not developed following a specific theoretical viewpoint and evaluates the patient’s in-therapy behaviour that may point to the existence of a therapeutic alliance: collaborative and active participation, spontaneous and full agreement, affective involvement, desire to achieve goals, confidence, clear, and realistic representation of the therapeutic relationship.	The report is filled in by the therapist and consists of six items rated on a five-point scale. Patients also respond to 12 items that rate the level of therapist collaboration.	Therapists	The scale has shown a high level of internal consistency (Cronbach’s alpha = 0.89), and has demonstrated good test–retest reliability during a 3-month period (average r = 0.72; Frank and Gundersen, 1990).	Alliance as measured by the PSR has been shown to be correlated with outcome in patients with severe and enduring mental illness such as schizophrenia (Elvins and Green, 2008; Svensson and Hansson, 1999).
Agnew relationship measure (ARM)(Agnew-Davies et al., 1998).	The ARM was intended to describe components of the alliance in language designed to be acceptable within a wide range of theoretical orientations and was developed during the Second Sheffield Psychotherapy Project, a randomized comparison of cognitive–behavioural therapy and psychodynamic–interpersonal therapy for depression. The ARM assesses five dimensions of the alliance: bond, partnership, confidence, openness, and client initiative.	The ARM has five scales comprising 28 items rated on parallel forms by patients and therapists using a seven-point scale.	Therapists/clients	Internal consistencies of four scales were all acceptable (alphas ranged from 0.77 to 0.87). The internal consistency of the Client Initiative scale was low (0.55; Agnew et al., 1998).	Some aspects of the alliance as measured by the ARM was correlated with psychotherapy outcome (Stiles et al., 1998). A qualitative study using an outcome measurement (CORE Net) in correlation with a modified five-point scale of the ARM confirms that the client’s rating of the alliance is one of the best predictor of engagement and outcome (Unsworth, 2008).
Kim alliance scale (KAS)(Kim et al., 2001)	The KAS was developed to rate the quality of the therapeutic alliance from the patient’s perspective. The scale comprises the three dimension of the alliance originally proposed by Bordin (1979) plus a fourth dimension: the patient’s empowerment, i.e., the patient becoming more responsible for his/her own care and more involved in making choices.	The KAS is a self report measure consisting of 30-item (8 collaboration item, 11 communication item, 5 integration item, and 6 empowerment item) each of one rated on a four-point scale.	Clients	High level of internal consistency for the total KAS (Cronbach’s alpha = 0.94). The alphas for the four dimensions ranged from 0.71 for empowerment to 0.87 for communication (Kim et al., 2001). Highly correlated with the ARM.	The scale has not been used in outcome research.

Any attempt to measure something as complex as therapeutic alliance involves a series of conceptual and methodological shortcomings, which have probably hindered the development of research in this field. Single-case research is one method used to investigate this theoretical construct, but implies some methodological drawbacks regarding the simultaneous treatment of several factors, the need for an adequate number of repeated measurements, and the generalizability of results. Meta-analysis is a possible research strategy that can be used to obtain the combined results of studies on the same topic. However, it is important to remember that meta-analysis is more valid when the effect being investigated is quite specific. According to Migone (1996), another hindrance is the so-called Rashomon effect (named after the 1950 film by Akira Kurosawa): each single aspect of therapeutic alliance may be perceived very differently by the therapist, patient, and clinical observer, which raises the question of objectivity.

Di Nuovo et al. (1998) propose some methodological changes to increase the utility of research findings, namely, omitting the use of methodological “control” techniques with comparisons between groups, re-evaluating single-case research, reconsidering the use of longitudinal studies, and using systematic replication and meta-analysis to guarantee the generalizability of results, even with single cases.

In spite of the difficulties involved in this type of research, Table 1 shows that numerous instruments have been developed to analyses the therapeutic alliance. Though designed by independent research teams, there is often good correlation between the scales used to rate the therapeutic alliance, which reveal that these instruments tend to assess the same underlying process (Martin et al., 2000). Fenton et al. (2001) compared the predictive validity of six instruments (CALPAS, Penn Scale, VTAS, WAI-Observer, WAI-therapist, WAI-Client) and found that all the measurement instruments used by raters (six trained clinicians served as independent raters for this study) were strong predictors of outcome. None of their findings suggest that any one instrument was a stronger predictor of outcome than the others, in relation to the type of therapy being considered.

It is interesting to note that although almost all of these scales were originally designed to examine the perspective of only one member of the patient–therapist–observer triad, they were later extended or modified to rate perspectives that were not previously considered. In short, some scales analyses specific theoretical concepts of the alliance (Penn scales, WAI, CALPAS, TBS), whereas others use a more eclectic construct (VPPS, VTAS, TARS). The number of items included in the scales varies considerably (between 6 and 145 items), as do the dimensions of the alliance investigated (e.g., two in the Penn scales; three in the WAI, TSR, and TBS; four in the CALPAS and KAS; and five in the ARM). According to Martin et al. (2000), the most frequently used scales in individual psychotherapy are the WAI, CALPAS, and Penn scales, followed by the Vanderbilt scales, TARS, and TBS.

Different approaches for the evaluation of alliance coexist in group psychotherapy. One of them is derived from individual psychotherapy. Johnson et al. (2005) used the WAI to refer to relationships with other group members; it was called the Member–Member WAI. The WAI-based scale used to measure relationships with group leaders was called the Member–Leader WAI. The CALPAS Group used by Crowe and Grenyer (2008) consisted of four subscales: patient working capacity, patient commitment, working strategy consensus, and member understanding and involvement.

Although a comparison between different treatment modalities is a topic beyond the scope of this paper, it is worth noting that in the late 1980s, some authors (Marmar et al., 1989a,b) failed to demonstrate significant differences between behavioral, cognitive, and brief psychodynamic therapies in the level of alliance as measured by CALPAS. However, subsequently, Raue et al. (1997), when comparing psychodynamic–interpersonal and cognitive–behavioral therapy sessions, found that observers rated the cognitive–behavioral group significantly higher on the WAI. This latter study compared 57 clients, diagnosed with major depression and receiving either psychodynamic–interpersonal or cognitive–behavioral therapy: the cognitive–behavioral sessions were rated as having better therapeutic alliances than the psychodynamic ones. They argue that these findings could reflect the effort in cognitive–behavioral therapy to give clients positive experiences and to emphasize positive coping strategies. A more recent comparison was suggested by Spinhoven et al. (2007), whose aim was to evaluate the therapeutic alliance in schema-focused therapy (Young et al., 2003; Nadort et al., 2009) and transference-focused psychotherapy (Yeomans et al., 2002). Results obtained by evaluating alliance through WAI-Client and WAI-therapist after 3, 15, and 33 months, showed clear alliance differences between treatments, suggesting that the quality of the alliance was affected by the nature of the treatment. Schema-focused therapy, with its emphasis on a nurturing and supportive attitude of therapist and the aim of developing mutual trust and positive regard, produced a better alliance according to the ratings of both therapists and patients. Ratings by therapists during early treatment, in particular, were predictive of dropout, whereas growth of the therapeutic alliance as experienced by patients during the first part of therapy, was seen to predict subsequent symptom reduction.

Phases of the Alliance during the Therapeutic Process and the Relationship with the Outcome

There is much debate on the role of the therapeutic alliance during the psychotherapeutic process. It may in fact be a simple effect of the temporal progression of the therapy rather than an important causal factor. On the basis of this hypothesis, we would expect a development in the alliance to be characterized by a linear growth pattern over the course of the therapy, and alliance ratings obtained in the early phases to be weaker predictors of outcome than those obtained toward the end of the therapy. However, according to the findings of numerous researchers, this is not the case. Safran et al. (1990) conclude that the positive outcome of therapy was more closely associated with the successful resolution of ruptures in the alliance than with a linear growth pattern as the therapy proceeds. Horvath and Marx (1991) describe the course of the alliance in successful therapies as a sequence of developments, breaches, and repairs. According to Horvath and Symonds (1991), the extent of the relationship between alliance and outcome was not a direct function of time: they find that measurements obtained during the earliest and most advanced counseling sessions were stronger predictors of outcome than those obtained during the middle phase of therapy.

The results of these studies have led researchers to consider the existence of two important phases in the alliance. The first phase coincides with the initial development of the alliance during the first five sessions of short-term therapy and peaks during the third session. During the first phase, adequate levels of collaboration and confidence are fostered, patient and therapist agree upon their goals, and the patient develops a certain degree of confidence in the procedures that constitute the framework of the therapy. In the second phase the therapist begins to challenge the patient’s dysfunctional thoughts, affects, and behavior patterns, with the intent of changing them. The patient may interpret the therapist’s more active intervention as a reduction in support and empathy, which may weaken or rupture the alliance. The deterioration in the relationship must be repaired if the therapy is to be successful.

This model implies that the alliance can be damaged at various times during the course of therapy and for different reasons. The effect on therapy differs, depending on when the difficulty arises. In the early phases, it may create problems in terms of the patient’s commitment to the process of therapy. In this case, the patient may prematurely terminate the therapy contract. In more advanced phases of therapy, an interruption in the alliance may be triggered by a number of therapeutic scenarios, including when patients’ thoughts and emotions have been invalidated in some way. Within a transference-focused psychotherapy framework, the patient’s expectations of the therapist may be unrealistic and idealized, which may therefore hinder their ability to use the therapy to deal with important issues. In situations such as this, the actual therapeutic alliance regularly and repetitively reflects the patient’s unresolved conflicts.

According to Safran and Segal (1990), many therapies are characterized by at least one or more ruptures in the alliance during the course of treatment. Randeau and Wampold (1991) analyses the verbal exchanges between therapist and patient pairs in high and low-level alliance situations and find that, in high-level alliance situations, patients responded to the therapist with sentences that reflected a high level of involvement, while in low-level alliance situations, patients adopted avoidance strategies. Although some studies are based on a very limited number of cases, the results appear consistent: the therapist’s focus on the patient’s conflictual behavior patterns and the patient’s involvement rather than avoidance in responding to these challenges, are factors that contribute to improving the therapeutic alliance. Fluctuations in the alliance, especially in the middle phase, thus appear to reflect the re-emergence of the patient’s dysfunctional avoidant strategies and the task of the therapist is to recognize and resolve these conflicts.

While recent theorists have stressed on the dynamic nature of the therapeutic alliance over time, most researchers have used static measures of alliance. There are currently several therapy models that consider the temporal dimension of the alliance, and these can be divided into two groups: the first comprises those addressing transitional fluctuations in alliance levels, while the second consists of those concerned with the more global dynamics of the development of the alliance.

Few studies have analyzed alliance at different stages in the treatment process. Hartley and Strupp (1983) examined ratings obtained during the first session and then during sessions representing 25, 50, 75, and 100% of the treatment, over the course of short-term therapies. Among patients who completed the therapy successfully, there was an increase in the alliance rating between the first session and the session representing the 25% mark, whereas among unsuccessful patients, the alliance rating declined over the same period. According to the results proposed by Tracey (1989), the more successful the outcome, the more curvilinear the pattern of client and therapist session satisfaction (high–low–high) over the course of treatment. When the outcome was worse, the curvilinear pattern was weaker.

Horvath et al. (1990) posit an initial phase in which the alliance was strong, followed by a period of decline, and a subsequent period of repair. Kivlighan and Shaughnessy (1995) use the hierarchical linear modeling method (an analysis technique for studying the process of change in studies where measurements are repeated) to analyses the development of the alliance in a large number of cases. According to their findings, some dyads presented the high–low–high pattern, others the opposite, and a third set of dyads had no specific pattern, although there appeared to be a generalized fluctuation in the alliance during the course of treatment.

In recent years, researchers have analyzed fluctuations in the alliance, in the quest to define patterns of therapeutic alliance development. Kivlighan and Shaughnessy (2000) distinguish three patterns of therapeutic alliance development: stable alliance, linear alliance growth, and quadratic or “U-shaped pattern” alliance growth. They based their analysis on the first four sessions of short-term therapy and focused their attention on the third pattern, in that this appeared to be correlated with the best therapeutic outcomes.

In further studies of this development pattern, Stiles et al. (2004) analyzed therapeutic alliance growth during the course of short-term treatment of depressed patients, drawn from the Second Sheffield Psychotherapy Project, who received cognitive–behavioral and psychodynamic–interpersonal therapy. Unlike Kivlighan and Shaughnessy, these authors considered therapies consisting of 8 and 16 sessions, using the ARM to rate the therapeutic bond, partnership, and confidence, disclosure, and patient initiative. Cluster analysis yielded four therapeutic alliance development patterns, two of which matched Kivlighan and Shaughnessy’s patterns: stable alliance; linear alliance growth with high variability between sessions; negative growth with high variability between sessions; and positive growth with low variability between sessions. No significant correlation was observed between any of the four patterns and the therapeutic outcome. However, the authors observed a cycle of therapeutic alliance rupture–repair events in all cases: very frequent ruptures followed by rapid resolution processes, that is, V-shaped patterns. On the basis of this characteristic, the authors hypothesize that the V-shaped alliance patterns may be correlated with positive outcomes. In particular, Stiles et al. (2004) provide the first statistical demonstration of the hypothesis previously formulated by Safran and Muran (2000) and Samstag et al. (2004), where the alliance ruptures represented opportunities for clients to learn about their problems relating to others, and repairs represented such opportunities having been taken in the here-and-now of the therapeutic relationship.

The results of the study by De Roten et al. (2004) produced two patterns of alliance development (linear and stable), but no quadratic (U-shaped) or rapid rupture–repair (V-shaped) patterns emerged. The authors provided a possible explanation for these results by attributing them to the type of psychotherapy being investigated (the Brief Psychodynamic Investigation proposed by Gilliéron, 1989, which is a manual on a very brief psychotherapeutic four-session intervention) and the type of sample (psychiatric patients). Moreover, a new rating scale, the HAq, had replaced those that were used previously (WAI and ARM). According to De Roten et al. (2004), these results were in line with Horvath’s view of the alliance as a constructive process, rather than with the views of Gelso and Carter (1994) and Safran and Muran (1996) concerning the rupture and repair of alliances, in which change was a better predictor of stability outcomes. De Roten et al. (2004) suggest that a process characterized by ruptures and repairs was more likely to occur in long-term psychodynamic treatment, particularly during phases of in-depth work.

According to Castonguay et al. (2006), patterns of therapeutic alliance development require further investigation, in order to understand how and whether the various patterns are a cause, effect, or manifestation of improvement. This has supported the idea that therapeutic alliance may be characterized by a variable pattern over the course of treatment, and led to the establishment of a number of research projects to study this phenomenon.

Discussion and Conclusion

According to their meta-analysis based on the results of 24 studies, Horvath and Symonds (1991) demonstrate the existence of a moderate but reliable association between good therapeutic alliance and positive therapeutic outcome. More recent meta-analyses of studies examining the linkage between alliance and outcomes in both adult and youth psychotherapy (Martin et al., 2000; Shirk and Karver, 2003; Karver et al., 2006) have confirmed these results and also indicated that the quality of the alliance was more predictive of positive outcome than the type of intervention (but for slightly different results in youth psychotherapy see McLeod, 2011).

Some theorists have defined the quality of the alliance as the “quintessential integrative variable” of a therapy (Wolfe and Goldfried, 1988), and in the present state, it seems possible to affirm that the quality of the client–therapist alliance is a consistent predictor of positive clinical outcome independent of the variety of psychotherapy approaches and outcome measures (Horvath and Bedi, 2002; Norcross, 2002). Thus, it is not by chance that in their meta-analysis, Horvath and Luborsky (1993) conclude that two main aspects of the alliance were measured by several scales regardless of the theoretical frameworks and the therapeutic models: personal attachments between therapist and patient, and collaboration and desire to invest in the therapeutic process.

In our opinion, regarding the relationship between the therapeutic alliance and the outcome of psychotherapy, future research should pay special attention to the comparison between patients’ and therapists’ assessments of the therapeutic alliance: these have often been found to differ, and evidence suggests that the patient’s assessment is a better predictor of the outcome of psychotherapy (Castonguay et al., 2006). In Horvath’s (2000) opinion, this might be explained by the limitations of assessment procedures, since the rating scales are usually validated on the basis of patient data, whereas the therapist views the relationship through a “theoretical lens,” thus tending to assess the relationship according to what the theory suggests is a good therapeutic relationship or according to the assumptions about the signs that indicate the presence or absence of the desirable relationship qualities. On the other hand, the patients’ assessments tend to be more subjective, atheoretical, and based on their own past experiences in similar situations. This accounts for the difficulties associated with the concept of alliance, which is built interactively, and so any assessment must also consider the mutual influence of the participants. In a helpful contribution, Hentschel (2005) points out that the problematic aspect of empirical studies investigating the alliance is their tendency to view the alliance construct as a treatment strategy and a predictor of therapeutic outcome: if the therapist is instructed, for instance, on methods of increasing the level of alliance, and is then asked to rate the alliance, this can lead to a contamination of the results. The use of neutral observers or the creation of counterintuitive studies is therefore recommended.

From this historical excursus, it is clear that research into the assessment of the psychotherapeutic process is alive and well. The development of a dynamic vision of the concept of therapeutic alliance is also apparent. The work of theorists and researchers has contributed toward enriching the definition of therapeutic alliance, first formulated in 1956. Research aimed at analyzing the components that make up the alliance continues to flourish and develop. Numerous rating scales have been designed to analyses and measure the therapeutic alliance, scales that have enabled us to gain a better understanding of the various aspects of the alliance and observe it from different perspectives: from that of the patient, therapist, and observer. Attention has recently turned toward the role of the therapeutic alliance in the various phases of therapy and the relationship between alliance and outcome.

So far, few studies have regarded long-term psychotherapy involving many counseling sessions. This topic, along with a more detailed examination of the relationship between the psychological disorder being treated and the therapeutic alliance, will be the subject of future research projects. Equally important, in our opinion, will be the findings of studies regarding drop-out and therapeutic alliance ruptures, which must necessarily consider the differences between that perceived by the patient and that perceived by the therapist.

Conflict of Interest Statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.