Send email for updates
Flückiger C., Del Re A.C., Wampold B.E. et al.
Psychotherapy: 2018, 55(4), p. 316–340.
Unable to obtain a copy by clicking title? Try asking the author for a reprint by adapting this prepared e-mail or by writing to Dr Flückiger at email@example.com. You could also try this alternative source.
Comprehensive review for the American Psychological Association concludes that the working relationship between clients and their counsellors or therapists is one of the largest and most consistent determinants of outcomes. Practice recommendations aim to help practitioners foster strong relationships.
Summary [Though not specific to clients with drug and alcohol problems, the principles derived from this review of psychotherapy studies are likely to be applicable, partly because severe substance use problems generally form part of a complex of broader psychosocial problems. This review updates an earlier version also in the Effectiveness Bank.]
Commissioned by a task force of the American Psychological Association, this review amalgamated findings relating outcomes to the strength of the collaborative working relationship and emotional bond (‘alliance’) between clients and therapists in the psychotherapy of individual clients rather than groups or families.
The link between alliance and outcomes was moderate and statistically significant. Assuming a causal connection, in the context of other influences it would be a relatively important determinant of patient progress.
Though causality cannot be established by the types of studies included in the analysis, it is probable, and the safest stance is to presume that how the therapist is and behaves affects how well their patients do, and does so partly via the alliance they help generate with the client.
The featured review is one of several in a special issue of the journal Psychotherapy devoted to features of the therapist-client relationship related to effectiveness, based on the work of a task force established by the American Psychological Association. This particular review examined the links between outcomes of individual psychotherapy and the alliance between therapist and client. It complements reviews on the alliance in therapy for couples and families and in child and adolescent therapy.
‘Alliance’ has been variously defined as a bond between the therapist and client which holds the client in therapy or as a collaborative working relationship, and is sometimes seen as mainly working at the unconscious level, sometimes at the conscious. An influential formulation sees it as a collaborative stance composed of: agreement between patient and therapist on the goals of therapy; agreement on the tasks to be undertaken during therapy; and an emotional bond between patient and therapist. Rather than a specific task or goal, the alliance infuses every interaction throughout psychotherapy, not just when the interaction focuses on the relationship or on agreeing goals and tasks.
In practice, in each study the alliance is defined by which of the diverse questionnaires or other methods are used to assess it. Four measures accounted for about two-thirds of the studies found by this review, and one – the Working Alliance Inventory – was the basis for 69% of the studies panel. Often these methods deliver varying assessments, so are clearly measuring somewhat different facets of the alliance, but at least three have been found to share the central common theme of a confident, collaborative relationship.
The Working Alliance Inventory completed by patient and/or therapist includes statements indicative of the three dimensions of the alliance mentioned above. Sample statements below are taken from a short version intended to be completed by clients, who respond by choosing options ranging from “seldom” to “always”.
“[My therapist] and I have established a good understanding of the kind of changes that would be good for me.”
“I feel that the things I do in therapy will help me to accomplish the changes that I want.”
“I feel that [my therapist] appreciates me.”
The review incorporated a meta-analysis which amalgamated results from relevant studies to provide estimates of the overall strength of the link between alliance and outcomes, and to be able to probe for influences on the strength of that link. The strength of the alliance–outcomes link was calculated as a correlation coefficient, an expression of the degree to which outcomes co-varied with the solidity of the alliance. The chosen metric ranges from -1 (perfect negative co-variation, meaning that as one side of the link gets larger the other diminishes) to +1 (perfect positive co-variation, meaning that as one side of the link gets larger so does the other). Correlation coefficients were also converted to effect sizes. Effectively these metrics indicate how influential the alliance had been if causally linked to outcomes.
The analyses included studies of adult patients receiving individual therapy in a clinical context (rather than for example, ‘pretend’ patients recruited for the study) either face-to-face or via computerised links, where the relationship of the alliance to outcomes was reported in a way in which it could be amalgamated with similar results from other studies. Usually the outcomes related to general mental health or the specific problems which brought the client to therapy, but sometimes also or instead to drop-out from therapy. Generally the alliance was measured from the client’s perspective. The search extended to reports in Italian, German or French, as well as English. The resulting 306 studies examined outcomes in 295 different samples totalling about 30,000 clients.
Overall the strength of the link between alliance and psychotherapy outcomes equated to a statistically significant correlation of 0.28 and a moderate-strength effect size of 0.58, an association which accounts for about 8% of the differences in outcomes across the samples of patients. In other words, the more solid the working relationship between therapist and client, the better outcomes tended to be, though not consistently.
The strength of the link varied between studies substantially more than would be expected by chance. Many possible reasons for this variation were explored, not all of which are reported here. There was no substantial or significant variation depending on the type of therapy (counselling, cognitive-behavioural, psychodynamic, humanistic, interpersonal, or unspecified/eclectic), reinforcing the view that the alliance is a ‘common factor’ (1 2) underpinning any effective psychosocial therapy.
However, the type of patient did seem to make a difference. Notably in the present context, at 0.14 the 29 studies of the treatment of substance use problems recorded a statistically significant correlation, but one significantly smaller than across all the studies, and smaller too than the treatment of other problems including anxiety disorders (correlation 0.24) or depression (0.26). At 0.15, studies of the treatment of eating disorders also registered a relatively small correlation. Neverthless, the alliance–outcomes link was significant within all the diagnostic categories tested and fairly stable at usually around 0.2 to 0.3. The relatively weaker link in the treatment of substance use may have been partly due to treatment drop-out being more often used as an outcome; perhaps because it has multiple causes, across all studies drop-out was relatively weakly related to the alliance. Another possibility is the fact that substance use patients were more often from ethnic minorities, among whom alliance may be less of a determinant of outcome.
Which questionnaire or other method was used to measure the alliance made no significant difference to the strength of the link, and nor did who rated the alliance (client versus therapist, observer, or other participant). In particular, the strength of the link was about the same whether assessed by client (0.25) or therapist (0.22). When alliance was assessed within three sessions of the end of therapy – closest to when outcomes were also assessed – at 0.30 the correlation with outcomes was stronger than when it was assessed earlier. However, at 0.22 the link remained significant even when alliance was measured in sessions one to five.
Patients are partners in forging the alliance, and there is some evidence that those with certain features like a trusting attitude and social support do so more readily. However, there is little evidence that the alliance–outcome correlation is systematically impacted by the patient’s characteristics at the start of treatment, and some evidence that this correlation is strongly affected by the therapist.
The featured review contradicted the common presumption that the alliance is less important in internet-based than in face-to-face therapies. Among the articles it found were 18 reporting on outcomes for 23 samples totalling 1,178 clients offered therapy via the internet, e-mail, video-conferencing, or over the phone. Across these studies, the strength of the link between alliance and psychotherapy outcomes equated to a statistically significant correlation of about 0.28 and a moderate-strength effect size of 0.57, very similar to findings for face-to-face psychotherapy.
That there is a close and possibly causal link between alliance and outcomes is reinforced by studies which found that session by session, changes in the alliance are followed by corresponding changes in the patient’s symptoms. These findings are consistent with, but not sufficient to prove, the theory that stronger alliances actually cause better outcomes. Direct evidence of a causal link would require studies which randomly allocated patients to therapeutic programmes which deliberately generated strong versus weak alliances. On ethical, conceptual and methodological grounds, such studies are not possible, meaning evidence for causality primarily relies on the accumulation of indirect and contextual evidence.
The featured review added to this evidence by helping to exclude an alternative explanation of the alliance–outcomes link – that rather than causing changes in the patient’s symptoms, the alliance merely reflects the severity of these symptoms at the start of treatment or early changes during therapy. If this was the case, then the alliance–outcomes link should be significantly weaker once it was adjusted for initial severity and/or early symptom change. Of the 306 studies included in the featured review, 66 adjusted the alliance–outcomes association for such possible influences. Across these studies, at 0.22 the amalgamated alliance–outcomes correlation after adjustment was only slightly and non-significantly weaker than the 0.25 before adjustment, supporting the hypothesis that the association between alliance and outcome is not primarily an epiphenomenon of patient characteristics and early therapy gains.
Studies of the alliance form among the richest bodies of research on psychotherapy process and outcome. Based on that research, the following practices are recommended:
• Build and maintain the alliance throughout therapy. That entails creating a warm emotional bond or collaborative attachment with the patient.
• Early on develop agreement on therapy goals and on the respective tasks of patient and practitioner. These agreements reliably predict therapeutic success.
• Agreement on goals and tasks does not mean patient or therapist automatically accepting the other’s formulations. A strong alliance is often a result of negotiation.
• Respond to the client’s motivation or readiness to change and ability to change during the early sessions of therapy.
• Customised to the patient, use language with the qualities of inclusiveness and negotiation, and collaborate not just in words but also via non-verbal communications.
• Directly and immediately address tensions or breakdowns in the alliance.
• Treat each patient as an individual by being responsive to their problems and preferences.
• Regularly assessing the client’s perspective on the strength or quality of the alliance helps detect unsatisfactory progress and impending drop-out.
• Therapist and client do not always agree on the state of the alliance. Divergences should be interpreted carefully, because they do not necessarily indicate disagreement, but perhaps different expectations or perspectives. Disagreement is not something negative, but instead may indicate that discussing the relationship could be helpful or necessary.
• Pay as much attention to the alliance in technology-mediated as in face-to-face psychotherapy.
commentary Though research findings are not definitive, for reasons explained below the safest stance for trainers, supervisors, therapists, counsellors, patients and clients, is to presume that a good working relationship is an important determinant of treatment success, and that nurturing, maintaining, and as needed, re-establishing such a relationship, are core tasks not just in psychosocial therapies, but in treatment generally. The recommendations in the featured review ( above), plus the partially complementary ones in its earlier version, aim to aid therapists and counsellors in those tasks.
The strength of the alliance–outcomes link was virtually identical to that found in the earlier version of this analysis, suggesting that further studies are unlikely to alter the picture, an implication reinforced by the fact that there was no evidence that studies missed by the review would affect its results. In the reviewers’ words, “Given the robust finding of the positive association between alliance and outcome, major changes in the association are not likely in the future.” They called for “research designs … that can test the causal impact of the alliance”, but acknowledged that the definitive test – a randomised trial which can eliminate competing explanations of the alliance–outcomes link – is not on the agenda. This means that we are at or near the point where further research will not settle the question of whether forging a strong alliance is an active ingredient in promoting desired changes – a question which matters, because the moderate effect size attached to the alliance is large in relation to that usually found in studies of substance use therapies. If alliance is a causal factor, its influence is relatively substantial and consistent across different types of therapy, including the counselling and cognitive-behavioural approaches common in substance use treatment.
The reviewers’ practice recommendations are based on the likelihood of a causal link between alliance and patients’ progress, which can be leveraged by the therapist to augment that progress. In other words, that how the therapist is and behaves affects how well their patients do, and does so partly via the collaborative working bond they help form between themselves and the client. This bond can be seen as the convergence or emergent result of the components also addressed by reviews (listed at the end of this analysis) commissioned by the same American Psychological Association task force, including empathy, repairing ruptures in the client–therapist relationship, demonstrating positive regard for the patient, and conveying the credibility of the therapy.
Given the nature of the studies which supported these recommendations, causality cannot be considered proven ( below), but for at least two reasons it seems likely. First is the consistency of the association between the strength of the alliance and outcomes. Though sometimes very small and non-significant, in only five of the 295 cases amalgamated by the review was this relationship negative. Second is the plausibility of the proposition that establishing a good working relationship will help keep patients in therapy and actively working with the therapist towards agreed therapeutic goals, and that this greater opportunity for therapy to work will often translate into it actually working better.
Additionally, there seems little or nothing to lose and possibly much to gain from establishing a good working relationship with clients, nothing to gain and possibly much to lose from failing to do so, and ethical considerations demand a positive attitude to troubled individuals who have come to you for help.
The focus in the analyses discussed so far has been on psychosocial therapies, but the relationship between helper–client alliance or allied variables on the one hand, and outcomes on the other, has been found to be about the same across six other ‘helping’ professions: medicine, nursing, social work, physical therapy, education, and neurology/rehabilitation.
Forging a strong working relationship with a client should not be confused with universally agreeing with their views or complying with their wishes. The reviewers argued that “Goal and task agreement does not mean that the therapist automatically accepts the patient’s goals and tasks or vice versa.” Indeed, there is evidence that experiencing a successful resolution of a tension or breakdown in the client–worker alliance is associated with better outcomes than tension-free therapy – as if the experience of being able to work through a relationship difficulty is more instructive or therapeutic than unbroken ‘plain sailing’. Repairing so-called ruptures in the client–worker alliance can be done more directly in the context of an overall strong relationship.
Related findings emerged from an unusually deep analysis of data from five US outpatient counselling centres. Surprisingly, substance use reductions were best sustained by clients of counsellors rated about average in terms of their clients’ experiences of working with them. Counsellors who had been relatively poor at striking up a close alliance had worse outcomes, but so too did those who had been especially good.
In this study counsellors were generally very good at generating positive relationships; it was only towards the very top of this range that outcomes started to worsen. The questionnaire on which this finding was based suggests that therapists whose clients scored them at these extreme levels might have focused too much on the client’s comfort, failing to develop change-promoting “discrepancy” when needed, perhaps not willing to generate some discomfort by highlighting how the patient’s actions contradict their self-image and values. Importantly, while scoring at the very top of this scale may not have been ideal, counsellors did not have to slip very far down before things start getting worse again; the findings were no carte blanche for neglecting alliance-building.
The featured review found a weaker but still statistically significant link between alliance and outcomes for substance use therapy, positing that this may partly be due to the use of drop-out as an outcome and the much greater than average presence of ethnic minority patients. Some of the same authors specifically examined these issues in a paper based largely on the previous version of the featured analysis. They found 94 studies of the alliance–outcomes link which reported the racial or ethnic mix of the caseload, of which 38 also reported the proportion diagnosed with substance use disorders and 16 were studies specific to substance use.
As in the featured analysis, at 0.18 the alliance–outcomes correlation was lower among these 16 studies than among studies not focused on substance users. Additionally, across all the studies, the greater the proportion of non-alcohol problem drug users in the sample, the weaker the link between alliance and outcomes, though this was not the case in relation to the proportion of problem drinkers.
The link also weakened the greater the proportion of ethnic minorities in the sample of patients, and even more so specifically with the proportion of African Americans. However, with the studies usually conducted in the USA, drug use and ethnic minority status overlapped so much that when the association of one with outcomes was taken into account, the association of the other became non-significant, making it unclear whether the focus on drug use or the ethnic origins of the sample was the decisive factor. In the USA, studied treatment populations are often overwhelmingly black, not the case in the UK, where the white population dominates. If racial minority status is the active factor, there is the clear possibility that in cultures like that of the UK there will be no weakening of the alliance–outcomes link among problem drug users. On the other hand, if minority status in the form of illegal drug use is the active factor, the same finding may emerge in the UK.
Another overlapping possibility is that it was the disproportionately low socioeconomic status of both drug-using and black clients which partially accounted for the link being weaker among both categories, or that it was the entire complex of stigmatised substance use, racial minority status and poverty, which accounted for the findings. This would explain why the influence of the alliance remained strong among typically more socially integrated problem drinkers, but not among the more socially excluded and poorer problem drug use treatment populations. With so few studies reporting the socioeconomic status of the clients, the analysis was unable to investigate this possibility, but it was one the reviewers took seriously: “It may also be true that the alliance-outcome correlation[s] in these studies were influenced and confounded with other sociocultural taboos that many of these clients likely face, such as living in precarious housing conditions, functioning under occupational uncertainty, lacking social support, as well as struggling with various drug-related legal concerns. Recognizing and addressing social classes and low-income issues such as social isolation, psychosocial stress, and powerlessness might help to build up trust in the therapeutic setting.”
Perhaps too, for people who feel they have reasons not to trust public services, the approach taken by the individual worker is less salient than whether the service as a whole seems to be unreservedly on their side. A study conducted in England in 2006 found clear relationships between the degree to which patients engaged with substance use treatment and organisational features such as team-working and mutual trust, whether the service fostered open communication between staff and was receptive to their ideas and concerns, was adequately resourced, and had a clear mission and programme. Like a more or less coherent, well organised department store, all these and other features funnelled to a head in the interaction between staff and ‘customer’, affecting whether that customer wanted to ‘stay and buy’, or preferred to move on and/or do without what they had come for.
The proposition that instead or as well, use of drop-out as an ‘outcome’ in studies of drug dependence treatment accounts for the weaker association of alliance with outcomes is contradicted by the findings of a review which focused on alliance and drug treatment, though a few of the samples consisted of problem drinkers. Published 13 years before the featured review, it was based on 18 studies rather than 29. Its findings have been explored in the Effectiveness Bank’s Drug Treatment Matrix, the major one being that therapeutic relationships were more consistently associated with engagement and retention than post-treatment outcomes. We suggested this might mean that stronger relationships make clients want to stay around, but do less to make them better – though for some treatments (especially those based on medications), retention is vital to their effectiveness. However, the earlier review did not amalgamate studies’ findings, so could not make a quantitative comparison of the strength of the alliance–outcomes link when the outcome was drop-out versus other outcomes. Where it agreed with the featured review’s findings across psychotherapy was that the link with outcomes was greater when alliance had been assessed late in treatment, closer to when outcomes were also assessed, but remained in some studies even when assessed early.
Though a causal link between alliance and outcomes is plausible, and it would be safe to assume its reality and probably unsafe not to, such a link could not be established by the types of studies included in the featured analysis. Generally these documented the development of client perceptions of the alliance during the course of therapy, and related these perceptions to outcomes. Such studies are generally unable to eliminate the possibility that (for example) patients who were going to do well in any event were more likely to cooperate with and feel positive about their therapists, or that therapists more capable of generating these feelings were also more competent in other ways. In these scenarios, alliance would remain associated with better outcomes, but not because it helped cause them. The reviewers acknowledged that it is not even enough to show that better outcomes reliably follow stronger alliances. As causality theorists have explained, “Thunder correlates with power outages, but thunder does not cause power outages. To distinguish causal from noncausal correlations, it is important to control for alternative causes.” Without effectively random allocation of patients to high- and low-alliance therapies, these “alternative causes” cannot completely be eliminated.
As they are added to the Effectiveness Bank, listed below will be analyses of the remaining reviews commissioned by the American Psychological Association task force.
Cohesion in group therapy
Treatment outcome expectations
Alliance in couple and family therapy
Alliance in child and adolescent therapy
Repairing ruptured alliances between therapists and clients
The ‘real relationship’
Therapist self-disclosure and ‘immediacy’
Thanks for their comments on this entry to David Skidmore based in England, former probation officer, addiction counsellor and regional manager with the National Treatment Agency for Substance Misuse. Commentators bear no responsibility for the text including the interpretations and any remaining errors.
Last revised 10 January 2019. First uploaded 11 December 2018
Give us your feedback on the site (two-minute survey)
Open Effectiveness Bank home page
Add your name to the mailing list to be alerted to new studies and other site updates
REVIEW 2018 Cohesion in group therapy: a meta-analysis
REVIEW 2018 Congruence/genuineness: a meta-analysis
REVIEW 2018 Alliance rupture repair: a meta-analysis