5
Historically, there has been a tension between those who believe that particular treatments for various disorders are effective because of the specific ingredients of the treatments and those who believe that therapeutic ingredients common to all (or most) treatments are responsible for the benefits of psychotherapy. So, in some ways, what makes psychotherapy work depends on who you ask. In this chapter, the evidence for specific ingredients is presented followed by the evidence for common factors.
In medicine, specificity is established in two ways. First, a medicine (or a procedure) is compared with a placebo in a double-blind, randomized control group design, which controls for the psychological factors, such as hope, expectancy, and relationship with a healer. If the medication is found to be superior to the placebo, then there is evidence that the specific ingredients of the medication are responsible for the benefits because the only difference between the active medication and the placebo conditions is the specific ingredient purported to be remedial for the medical condition. The second way that specificity is established in medicine is to verify a system-specific sequence (Wampold, 2007; Wampold & Imel, 2015). According to Wampold (2007), this sequence is accomplished in the following way:
(a) a biological explanation for the illness, based on scientific research, is established; (b) a treatment is designed or a substance is hypothesized to remediate the biological deficit; (c) administration of the substance demonstrably alters the biology of the patient in the expected way, and other substances do not; and (d) the change in the biology remediates the illness (a cure or management of chronic illness). (p. 867)
An example of a systemic-specific sequence involves the treatment of peptic ulcers. Peptic ulcers are hypothesized to be caused by the presence of a significant population of the bacterium Helicobacter pylori, the treatment consists (in part) of the administration of an antibiotic designed to reduce the population of the bacteria, the administration of the treatment does indeed reduce the population of bacteria, and the client subsequently improves (i.e., symptoms disappear, and tests reveal that the ulcer has healed).
To attempt to establish specificity in psychotherapy, psychotherapy researchers have used various experimental designs similar to the randomized placebo design and various other types of designs to establish the system-specific sequence. In many ways, establishing specificity is more difficult in psychotherapy than in medicine because of inherent difficulties in measuring process and outcomes in psychological systems, because double-blinding is impossible in psychotherapy trials and because the psychological deficits in various mental disorders are ambiguous. Nevertheless, in the next section, a brief review of the evidence for specificity, organized by type of design, is presented.
Component designs attempt to isolate the effects of particular ingredients by either removing a critical ingredient (often called a dismantling design), which should attenuate the benefits of the treatment, or by adding an ingredient (often called additive designs), which is hypothesized to increase the potency of the treatment (Borkovec, 1990). Two examples of component designs follow that illustrate the logic and provide prototypical results.
Cognitive behavior therapy (CBT) for depression, undoubtedly the most studied and validated psychotherapy, comprises three primary components: behavioral activation, acquisition of skills to interrupt and change automatic thoughts, and modification of core schema. Jacobson et al. (1996), in an attempt to establish the importance of the cognitive components, randomly assigned 150 depressed clients to one of three conditions: (a) behavioral activation only (BA condition), (b) behavioral activation plus ingredients purported to change automatic thoughts (AT condition), or (c) the full package (CBT), which also addresses the core cognitive schema. At the end of treatment and at 6-month follow-up, CBT was not more effective than either the BA condition or the AT condition. Moreover, all conditions changed automatic thoughts and dysfunctional attributional styles to the same degree. Those results suggest that the cognitive components of CBT are not specific for the treatment of depression; that is, the work on automatic thoughts and dysfunctional core attributions are not critical to the success of CBT for depression.
Resick et al. (2008) used a similar strategy to dismantle cognitive processing therapy (CPT) for posttraumatic stress disorder (PTSD) for female victims of interpersonal violence. The full CPT was compared with treatments that comprised one of the two main components of CPT: (a) written accounts of the traumatic violence (WA) or (b) cognitive therapy without the written accounts (CT). Clients in all three conditions improved substantially on measures of PTSD symptoms and on depression. The full treatment (i.e., CPT) was not significantly better than either of the two components, although CT was slightly superior to WA on PTSD measures. Similar to Jacobson et al. (1996), the full treatment was not superior to treatments containing some of the components, which suggests that the components were not specific for treating PTSD.
In 2001, Ahn and Wampold meta-analyzed all studies that used component designs to assess the specific ingredients. They were able to locate 27 studies that either removed a critical component or added a component to determine whether the full treatment package was superior to a treatment with fewer (or no) critical components. The effect for the difference between a treatment with and without the critical components was not significantly different from zero, indicating that across all of those studies, there was not a consistent demonstration that the hypothesized critical ingredients of treatments were critical to the success of the treatment.
In 2013, Bell, Marcus, and Goodlad (2013) replicated and extended Ahn and Wampold’s (2001) meta-analysis for component studies by (a) including more recent studies, (b) segregating additive and dismantling designs, (c) examining outcomes at termination and at follow-up, and (d) performing analyses on targeted versus nontargeted outcome measures. Informatively, no statistically significant effects for removing critical ingredients (i.e., dismantling designs) were found for targeted or nontargeted outcomes at either termination or follow-up. The only statistically significant effects were found for targeted outcomes of additive designs at both termination and follow-up.
In a meta-analysis of social anxiety, Acarturk, Cuijpers, van Straten, and de Graaf (2009) examined various treatments that contained various hypothesized ingredients. Although they were not component studies, they addressed the same issue. In that meta-analysis, treatments without exposure were not significantly different from those with exposure, treatments without relaxation were not significantly different from those with relaxation, and those without social skills training were not significantly different from those with social skills training.
Shortly after the development of the placebo control group design in medicine, Rosenthal and Frank (1956) suggested it be used in psychotherapy research to establish the specificity of psychological treatments. Unfortunately, use of the placebo control group design in psychotherapy is problematic in ways that demonstrate the differences between psychotherapy and medicine (Baskin, Tierney, Minami, & Wampold, 2003; Critelli & Neumann, 1984; Grünbaum, 1981; Kirsch, Wampold, & Kelley, 2016; A. K. Shapiro & Morris, 1978; Shepherd, 1993; Wampold, 1997, 2001a, 2001b; Wampold, Frost, & Yulish, 2016). The basic design of a psychological placebo has involved the creation of a treatment without any specific ingredients—that is, an inert psychological process. These placebo-type treatments often are called supportive counseling, alternative treatments, psychological placebos, supportive counseling, and common factor controls. The treatments have no psychological basis and often consist of empathic responding or reflective listening but no focus on particular problems, no coping skills, no cogent rationale for the treatment, and so forth. Sometimes, to control for certain specific ingredients, therapists are proscribed from responding in certain common ways. For example, to control for exposure in the treatment of PTSD, the placebo-type controls may require that the therapists not discuss the traumatic event, because that would be a form of covert or imaginal exposure (e.g., thinking about the event in a safe and comfortable environment; see, e.g., Foa, Rothbaum, Riggs, & Murdock, 1991).
Three primary problems exist with placebo groups in psychotherapy research. First is the problem of distinguishability. In medicine, the pill placebo and the active pill are indistinguishable, whereas in psychotherapy, they are quite different; that is, the active treatment and the placebo differ on many dimensions besides simply inclusion of the active ingredient, including the credibility of the treatment to the clients and the expectations that might be created by the actions. The best-designed placebos attempt to create a credible rationale for the treatment (Borkovec & Costello, 1993). On the other hand, placebos often are quite deficient, delivering a smaller dose of therapy or using less skilled therapists (Baskin et al., 2003). A second related problem is that placebo psychotherapy studies cannot be blinded. Clients may not know much about the various treatments in a study, but they certainly are aware of the ingredients of each condition because they are receiving them. More problematic, however, is that the therapists in the study are not blinded; typically, they know they are either providing an experimental, novel, and innovative treatment or a therapy without any active ingredients. The third problem is that the placebo control groups do not contain all of the common factors, particularly a cogent rationale and a treatment, two of the most critical common factors in many models, as was discussed in the previous chapters (Frank & Frank, 1991; Garfield, 1992; Imel & Wampold, 2008; Torrey, 1972; Wampold, 2007; Wampold, Imel, Bhati, & Johnson-Jennings, 2006; Yulish et al., 2017).
Despite the problems with placebo-type control groups, the evidence produced from designs with such control groups has been used to make claims about how psychotherapy works. In 1994, Lambert and Bergin (1994) reviewed 15 meta-analyses to determine the effects of treatment and psychological placebos. They concluded that treatments intended to be therapeutic were superior to psychological placebos (effect size = .48) and that psychological placebos were superior to no treatment (effect size = .42), both medium-sized effects (Cohen, 1988). Advocates of specificity might claim that this conclusion is supportive of their position because treatments with specific ingredients outperformed treatments without ingredients. On the other hand, common factor advocates would say that the controls do not contain all the common factors (e.g., a rationale and therapeutic actions), and given that a condition without a rationale and treatment can produce a sizable effect is evidence for the power of some of the common factors (Wampold & Imel, 2015; Yulish et al., 2017). Moreover, the difference between treatment with a cogent rationale and treatment versus one without could be interpreted as the degree to which a treatment (i.e., any cogent and convincing treatment delivered by a therapist who intends the treatment to be therapeutic) is important.
Additional meta-analyses using placebo-type control groups have attempted to address these issues to establish specificity. Stevens, Hynan, and Allen (2000) included only studies that compared three groups: (a) a specific treatment, (b) a placebo-type control group (which they labeled as a common factor control group), and (c) a no-treatment control group. They also attempted to control for the credibility of the placebo-type control group. They found larger differences between the specific treatment and the placebo-type control group and smaller differences between the placebo-type control and no-treatment groups than did Lambert and Bergin (1994), results that became more apparent as severity increased. They reached the following conclusions:
In contrast to Klein’s (1996) claim that nothing specific is occurring in psychotherapy, we found evidence that the specific components of psychotherapy exceed common factor effects, as represented by the common factor controls in this meta-analysis. . . . [emphasis added; p. 283] Our meta-analysis clearly indicates that the specific components of psychotherapy exert a beneficial influence over and above the common factors delivered, especially for participants with more severe problems. (Stevens et al., 2000, p. 286)
However, as noted, the conclusion about specific effects in psychotherapy made from placebo-type controls is tempered by the fact that, not infrequently, the placebo-type controls contain only some of the common factors thought necessary and that often the common factor controls are not well designed. Baskin et al. (2003) sought to investigate this very issue by classifying placebo-type controls into those that were structurally equivalent to the active treatment and those that were not structurally equivalent in terms of therapy duration, format, and therapist training. They found that treatments were superior to structurally inequivalent placebo-type control groups but were not superior to well-designed placebo-type control groups. That is, when the structure of placebos is approximately the same as the structure of the active treatment, their effects were similar to those of the active treatments, providing some evidence that the claims of specificity may be due to aspect of the design of placebo-type control groups.
What is remarkable is that treatments without any specific ingredients are effective. Smits and Hofmann (2009) meta-analyzed the outcomes of 19 control treatments for anxiety and found them to produce “significant improvement” (p. 229) with low rates of attrition. In a meta-analytic review of 18 studies, Honyashiki et al. (2014) found that CBT was only marginally more effective than psychological placebos, which were significantly more effective than no treatment. Yulish et al. (2017), as mentioned previously, found that treatments involving a focus on the client’s problems (e.g., they provide an explanation for the client’s distress, have a rationale for overcoming the client’s problems or achieving the client’s goals, and contain therapeutic actions consistent with the rationale) have a greater benefit, particularly in reducing symptoms, than do treatments without such a focus. Thus, it appears that treatments without specific ingredients, but which involve a warm, empathic, and supportive therapist, are effective, but treatments that include warmth, empathy, and support, as well as a focus on the client’s problems, are even more effective.
It has long been recognized that establishing specificity in psychotherapy requires more than a demonstration that active treatments are more effective than placebo-type controls (Rosenthal & Frank, 1956), a point that is made more vital given the ambiguity of placebo-type control studies in psychotherapy. Unfortunately, establishing the system-specific sequence in psychotherapy is tortuously difficult. One of the particular difficulties is that the psychological deficits of mental disorders have not been unambiguously established. The classification systems for mental disorders (e.g., the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders; American Psychiatric Association, 2013) are objective-descriptive rather than etiological (Widiger & Trull, 2007; Zachar, 2000, 2015); that is, mental disorders are determined by patterns of symptoms without establishing particular psychological deficits (see Borsboom, 2017). The problem this creates for establishing specificity is illustrated by considering panic disorders. The six best explanations for panic either have been falsified or are not falsifiable (Roth, Wilhelm, & Pettit, 2005); therefore, attempts to establish that treatments are successful by modifying identified deficits underlying panic are impossible. The lack of established etiological pathways for mental disorders is quite different, say, from the earlier example in which it is known that the bacterium H. pylori causes peptic ulcers. Despite these problems, there have been many creative efforts to investigate the system-specific sequence in psychotherapy, as described here briefly.
A primary way to establish a system-specific sequence is to examine mediating constructs (Kazdin, 2009). If Treatment A targets System A to treat a disorder and Treatment B targets System B to treat the same disorder, then both Treatment A and Treatment B may be specific through different systems, and both benefit the client. For example, CBT for depression would focus on cognitive distortions and maladaptive attributions, whereas interpersonal therapy (IPT) for depression would focus on relationships and social support. This is an explanation for why there may be generally equivalent outcomes of psychotherapy (i.e., the dodo bird conclusions) and, at the same time, specificity of the various treatments through different systems (Wampold & Imel, 2015). Investigations could profitably determine the mediating effects of the treatments on their respective systems (i.e., CBT would change cognitions, whereas IPT would change social relationships). Of course, the designs get complex and the results are difficult to interpret at times, as the following examples illustrate.
A good example of the examination of mediating systems is the dismantling of CBT for depression by Jacobson et al. (1996), as discussed previously. In addition to examining the outcomes of components of CBT (i.e., the full package, AT condition, and BA condition), they also examined two mediating variables, negative thinking and dysfunctional attributions, that should have been altered as a result of the cognitive components (CBT and, to a lesser degree, the AT condition). Nevertheless, all three treatments, including the BA condition, which contained no cognitive interventions, changed negative thinking and dysfunctional attributions equally. That cognitive variables did not mediate the treatment and depression as expected suggests that CBT is not specific for depression.
In the area of anxiety, Anholt et al. (2008) examined the process of change in CBT and response prevention/exposure for obsessive-compulsive disorder (OCD), two treatments that have been shown to be effective for OCD. Theoretically, response prevention/exposure is a behavioral treatment focused on the compulsive aspects of OCD, whereas CBT is a cognitive treatment focused on the obsessive aspects. Anholt and colleagues hypothesized that that response prevention/exposure would reduce compulsions initially, which would be followed by a reduction in obsessions, whereas CBT would exhibit the opposite pattern. Contrary to predictions, the process of change in both groups was the same: Compulsions changed first in both treatments, and change in compulsions was the better predictor of final outcome.
Another salient example of a failure to show mediating effects is found in an analysis of the National Institute of Mental Health (NIMH) treatment of depression collaborative research program (Elkin, Parloff, Hadley, & Autry, 1985; Elkin et al., 1989). The two psychotherapy arms, CBT and IPT, were expected to alter dysfunction attitudes and social adjustment, respectively. However, the expected mediating relationship was not present:
Despite different theoretical rationales, distinctive therapeutic procedures, and presumed differences in treatment processes, none of the therapies produced clear and consistent effects at termination of acute treatment on measures related to its theoretical origins. This conclusion applies, somewhat surprisingly, not only to the two psychotherapies but also to pharmacotherapy as practiced in the [NIMH treatment of depression collaborative research program]. (Imber et al., 1990, p. 357)
It appears that these early studies did not produce the anticipated mediating effects that are predicted theoretically, a conclusion corroborated by a meta-analysis of studies that included cognitive mediating variables (Oei & Free, 1995). The analysis of the NIMH program found that, consistent with specificity hypothesis, the cognitive variables did mediate CBT and outcome. However, findings also showed that cognitive variables mediated noncognitive treatments and even pharmacotherapies; that is, the cognitive variables mediated the treatment and outcome in a way that was indistinguishable from that of CBT.
Other attempts to find specific effects, particularly for CBT, using mediational-type designs, however, have produced some encouraging evidence for specificity. Research in the area of what has been called sudden gains (Tang & DeRubeis, 1999; Tang, DeRubeis, Beberman, & Pham, 2005; Tang, DeRubeis, Hollon, Amsterdam, & Shelton, 2007; Tang, Luborsky, & Andrusyna, 2002) has produced evidence of specificity. A sudden gain is a large change in symptoms from one session to the next. In the treatment of depression, clients who showed dramatic change in the Beck Depression Inventory (Beck, Ward, Mendelson, Mock, & Erbaugh, 1961) between sessions were functioning better at the end of treatment and at various intervals after treatment, and, furthermore, had fewer relapses (Tang & DeRubeis, 1999; Tang, DeRubeis, Hollon, et al., 2007; Tang, Luborsky, & Andrusyna, 2002). Of importance to specificity is that Tang et al. (2005) found that sudden gains in CBT for depression were preceded by substantial cognitive changes in the sessions before the sudden gains. Sudden gains have been found in other treatments (e.g., support-expressive psychotherapy), but the gains seemed to be less stable and not as related to outcome as they were in CBT (Tang et al., 2002). Strunk, Cooper, Ryan, DeRubeis, and Hollon (2012) found that the skills acquired in CT for depression were related to rates of relapse. Specifically, among clients who responded to CT, acquisition of cognitive coping strategies and in-session evidence of the use of cognitive skills were associated with lower rates of relapse 1 year later, even accounting for symptom level at the end of treatment or changes in symptoms over the course of therapy. This finding is informative because it indicates that specific skills learned during therapy may be important for the maintenance of the benefits of psychotherapy.
Lately, researchers have been disaggregating data collected over the course of therapy into components that are between clients (i.e., how does this client’s value on a variable compare with other clients) and within clients (i.e., how the client’s value on a variable at a given time point—say, a particular session—compares with the value on this variable at other time points; Curran & Bauer, 2011). The within-client effects are particularly informative about mediating effects because one would expect that a particularly process would occur before symptom reduction in particular therapies. Johnson, Hoffart, Nordahl, and Wampold (2017) recently compared metacognitive therapy (MCT) for various anxiety disorders to CBT designed for each particular anxiety disorder and found that MCT was moderately more effective at the end of treatment but that, at follow-up, there were no differences. It was expected that MCT would modify metacognitions, which then would result in a reduction of anxiety symptoms, whereas CBT would modify cognitions, which then would result in a reduction in anxiety. However, both CBT and MCT reduced both cognitions and metacognitions, which subsequently led to a reduction in anxiety (Johnson et al., 2018).
It may be that changes in the mediating system are unrelated to final outcome, which might be due to the common factors, but are related to longer term outcomes. An argument could be made that fundamental changes in a psychological system provide the client the bases with which to cope with various problems that will occur in the client’s life after therapy. The common factors may be sufficient to provide immediate relief, perhaps through a remoralizing process, as suggested by Jerome Frank (Frank & Frank, 1991), but few lasting changes are made. There is some evidence of this effect. In a study of 35 clients who responded to CBT for depression, it was found that those clients who acquired cognitive coping skills and displayed evidence that they were using CBT principles had lower relapse rates during the year following treatment (Strunk, DeRubeis, Chiu, & Alvarez, 2007). On the other hand, change in self-esteem was not related to relapse rates.
One of the explanations for the appearance of dodo bird results (i.e., general equivalence of outcomes) is that a given disorder may have multiple etiological pathways. For example, the cause of depression may vary; some clients’ depression may be caused by dysfunctional cognitions, for others, poor social relations; and for still others, a biological deficit or condition. This is essentially a criticism of diagnostic categories as descriptive rather than etiological (see Follette & Houts, 1996). If multiple etiological pathways exist for a given disorder, then a treatment designed for one particular deficit purportedly will be effective only for those clients who have the identified deficit; other treatments, aimed at other deficits, will be effective with other types of clients. Essentially, this is a matching hypothesis: Treatments designed to remediate a particular deficit will be effective when matched with clients with that deficit.
A number of studies have attempted to test the deficit matching hypothesis. In the late 1980s, Dance and Neufeld (1988) reviewed matching studies and found little evidence that matching a treatment to clients with a particular deficit provided produced more benefit. Similarly, B. Smith and Sechrest (1991) claimed that the evidence for this type of matching was “discouraging” (p. 237). Two multisite trials in the area of alcohol use disorders were designed to specifically test matching hypotheses, and neither one corroborated any of the matching hypotheses (Project Match Research Group, 1997; UKATT Research Team, 2008). For example, in the UKATT study (UKATT Research Team, 2008), it was hypothesized that clients who were not as ready to change would have better outcomes with motivational enhancement therapy, whereas those with low levels of social support would benefit from social behavior and network therapy. However, the results of the outcomes with more than 700 clients found no support for this hypothesis. Wampold and Imel (2015) reviewed the matching studies and found little evidence to corroborate that treatments matched to psychological deficits are more effective than treatments that are not matched.
It should be recognized that finding evidence for specific effects in psychotherapy is a difficult task—much more difficult than in medicine. For example, designing a placebo for a medication that is indistinguishable from the active medication and designing a double-blind, placebo trial is feasible, whereas, as discussed earlier, there inevitably are many issues in the comparable design in psychotherapy that lead to threats to the validity of the placebo design in psychotherapy. Similarly, research in the system-specific sequence in psychotherapy is more complex than in medicine because the causes of mental illness are more ambiguous, relatively speaking, because psychological systems are more difficult to observe and study than are biological systems.
Despite the difficulties with research on the issue of specificity, the evidence for specific effects in psychotherapy is relatively sparse. However, because of these difficulties, it is too early to close the door on specific effects. The unambiguous detection of specific effects for a particular disorder could lead to more efficient and effective services.
The paucity of evidence for specific factors leads to the question about whether the common factors are responsible for the benefits of psychotherapy. The complexity of making inferences from research on the common factors likely is greater than it is for the question of specificity. Fundamentally, for the most part the common factors cannot be experimentally manipulated, so attributions of causation are more contorted than they are in experimental designs. For example, one cannot assign a case to either have a good working alliance or not; the alliance is a variable that is assessed and then correlated with outcome. Thus, does a correlation of alliance and outcome imply that alliance causes outcome, or might outcome cause the therapist and the client to indicate a stronger alliance? Or might a third variable cause both alliance and outcome? And is the alliance-outcome relationship due to the therapist’s contribution to the alliance or to the client’s contribution to the alliance? That the alliance cannot be manipulated experimentally does not imply that the alliance cannot be causally related to good outcomes; it only means that there are a plethora of threats to the validity of conclusions about causation, as is discussed in this section.
A second difficulty is that the common factors are not discrete ingredients of therapy that can be added to a psychotherapy recipe to create a fine dish. The alliance, which involves agreement about the tasks of therapy, is necessarily tied to the delivery of a treatment. The person of the therapist is a participant in the relationship. The acceptability of the rationale of a treatment (i.e., an explanation of the client’s difficulty) depends both on the alliance and helps create the alliance. Can one have a strong alliance with a relatively nonempathetic therapist? The common factors mutually influence each other over time and thus form a complex system that is difficult to understand, let alone research. Nevertheless, the research evidence for several common factors indicates that they are critically important to the outcome of psychotherapy.
Despite the interconnected nature of the common factors, the research evidence for a select few common factors is reviewed here. This is important evidence because it supports the idea that how a treatment is delivered is more important than which treatment is delivered. It appears that several therapeutic factors are important across the wide variety of existing therapeutic approaches.
The concept of the working alliance originated in psychodynamic theory, but, in the 1970s, the idea became pantheoretical (Horvath & Luborsky, 1993). The modal model of the working alliance (or simply, the alliance) consists of three components: the bond between the therapist and the client, agreement about the goals of therapy, and agreement about the tasks of therapy (Hatcher & Barends, 2006; Horvath & Bedi, 2002; Horvath & Luborsky, 1993). The alliance has been characterized as “the participants’ [i.e., therapist and client] collaborative, purposive work” (Hatcher & Barends, 2006, p. 292). The interrelatedness of the alliance and other factors is readily apparent by the agreement about the goals and the tasks of therapy (Hatcher & Barends, 2006; Tryon & Winograd, 2002; Wampold, 2007); there can be no agreement about the goals and tasks of therapy without a treatment structure. In the research that uses treatments involving minimal responding and containing no treatment rationale or structure, the alliance must be composed entirely of bond because there are no goals and tasks. Hatcher and Barends (2006) succinctly made this point:
Alliance is a property of all components of therapy, a concept superordinate to these components and not a component itself. Viewing technique and alliance as equivalent components of therapy confuses 2 levels of thinking, as does conflating alliance with the overall therapy relationship. (p. 292)
Despite the theoretical subtleties of the alliance, a predominant question in psychotherapy research has focused on the magnitude of the relationship between the alliance and outcome. The modal design in this area has examined the correlation of the alliance, which is measured early in psychotherapy (around the third session), and the final outcome or change in functioning from pretreatment to termination. Several reviews and meta-analyses that have been conducted over the years have found a modest but consistent correlation in the range of .25 (Flückiger, Del Re, Wampold, & Horvath, 2018; Horvath & Bedi, 2002; Horvath, Del Re, Flückiger, & Symonds, 2011; Horvath & Symonds, 1991; Martin, Garske, & Davis, 2000). The higher the alliance is at an early session, the better the outcome. Except for initial severity of the client, there is no other variable that has been assessed early in therapy that predicts final outcome better than the alliance.
A number of results accentuate the importance of the alliance. First, the alliance is related to outcome, regardless of the type of treatment psychotherapy being offered (Carroll, Nich, & Rounsaville, 1997; Flückiger, Del Re, Wampold, Symonds, & Horvath, 2012; Horvath & Bedi, 2002; Krupnick et al., 1994, 1996; Wampold & Imel, 2015), contrary to an expectation that the alliance might be more important in treatments that emphasize the relationship (e.g., humanistic or psychodynamic) over those that are more structured (e.g., CBT). Indeed, evidence indicates that the alliance is related to outcome in psychopharmacological treatments (Blatt, Zuroff, Quinlan, & Pilkonis, 1996; Krupnick, Elkin, et al., 1994; Krupnick, Sotsky, et al., 1996). In addition, it does not make much difference whether the alliance is rated by therapists, clients, or observers, although slightly higher alliances have been found when clients rate the alliance (Horvath et al., 2011).
There are some difficulties in interpreting the correlation of alliance and outcome (Crits-Christoph, Gibbons, & Hearon, 2006; DeRubeis, Brotman, & Gibbons, 2005). The central issue is whether it can be concluded that the alliance caused the better outcomes. The first problem is that early treatment gains may cause better alliances, and thus it is the early gains in treatment that are creating the correlation between alliance and final outcome. Using a variety of complex statistical means, several attempts have been made to rule out the alliance as a consequence of early gains hypothesis. Many studies have found that the early gains are not a confound; that is, the alliance predicts outcome over and above early gains (Baldwin, Wampold, & Imel, 2007; Falkenström, Granström, & Holmqvist, 2013, 2014; Hoffart, Øktedalen, Langkaas, & Wampold, 2013; Klein et al., 2003; Zuroff & Blatt, 2006). However, some evidence indicates that the alliance is a product rather than a cause of outcome (DeRubeis & Feeley, 1990; Feeley, DeRubeis, & Gelfand, 1999; Strunk et al., 2012), and thus this matter is not settled (see Wampold & Imel, 2015).
A second threat to the alliance-outcome connection is that the correlation may be due to the client’s contribution to the alliance. Some clients present to therapy with a greater capacity to form interpersonal relationships, perhaps because they have a better attachment history (Mallinckrodt, 1991) and, consequently, form better alliances with their therapists. And it could be that these clients make better use of therapy and benefit more from it. In this scenario, the client’s contribution to the alliance creates better outcomes, regardless of the therapist and the therapy. On the other hand, it may be the therapist’s contributions that are important. The therapists who generally form better alliances across a range of clients may be those who produce better outcomes across a range of clients; that is, effective therapists are effective because they are better able to form alliances with their clients. Of course, it may be an interactive effect: Some therapists form better alliances with some types of clients, and when there is a good match, better outcomes result. Disentangling the various sources of the alliance (i.e., clients, therapists, and interactions), Baldwin et al. (2007) found that therapist contributions to the alliance were predictive of outcome, whereas client contributions and the interaction of clients and therapist were not predictive. That is, the therapists who generally formed better alliances with their clients also had better outcomes. However, for any given therapists, variability in the alliance from one client to another did not predict outcome (see also Trepka, Rees, Shapiro, Hardy, & Barkham, 2004). A meta-analysis of alliance studies confirmed the importance of therapist contributions to the alliance (Del Re, Flückiger, Horvath, Symonds, & Wampold, 2012). In addition, evidence indicates that training therapists relative to creating an alliance and working through ruptures in the alliance are effective (Crits-Christoph, Connolly Gibbons, et al., 2006; Hilsenroth, Ackerman, Clemence, Strassle, & Handler, 2002; Safran & Muran, 2000; Safran, Muran, Samstag, & Stevens, 2002). Those results are consistent with the discussion in Chapter 3 of this volume: It is critical that therapists elicit collaboration with the client with regard to the explanation provided (i.e., the rationale of therapy) and the tasks of therapy. Thus, the theoretical model, which provides both rationale and tasks, is central to making therapy work.
Although it is difficult to make interpretations about the alliance in therapy for the reasons discussed, there is agreement that it is a critical aspect of all therapy. Even staunch supporters of the specificity in psychotherapy would admit that a good alliance is necessary, although not sufficient, for therapeutic change (see, e.g., Hofmann & Barlow, 2014). The evidence seems to indicate that alliance is not due to prior change, although some disagree. Nevertheless, the therapist’s contribution to the alliance is what is important with regard to producing better outcomes for that client.
One of the tenets of the common factor models of psychotherapy is that the person of the therapist is a critical ingredient (Wampold, 2007; Wampold & Imel, 2015). According to this view, some therapists, regardless of the approach undertaken, are more effective than others. The central question here is, Do some therapists consistently produce better outcomes than other therapists? And, if so, What are the characteristics and actions of effective therapists? Historically, the provider of services in many contexts has been ignored because the emphasis has been on the treatment or program (Danziger, 1990; Wampold & Imel, 2015). For example, generally in education, research has been aimed at innovations in curriculum or school reform, rather than on the teacher who is providing the curriculum (Nye, Konstantopoulos, & Hedges, 2004). Research in agriculture, an area that has spawned many of the statistical methods used in psychology, has been interested in farming practices (e.g., crop varieties, irrigation, fertilizers) rather than variations in implementation of those practices by the farmers themselves. In medicine, the drug or procedure is paramount, and the differences in outcomes among physicians have been ignored. Psychotherapy, particularly in the past few decades, has similarly ignored the therapist who delivers the approach (Beutler et al., 2004). Advances in statistical methods (i.e., multilevel modeling) have allowed more precise estimates of therapist effects by taking into consideration that clients are nested within therapists (Snijders & Bosker, 1999).
The first source of evidence with regard to therapist effects comes from clinical trials. Because therapists have been ignored, rarely, if ever, are clinical trials designed to detect therapist effects. Nevertheless, a number of reanalyses have been conducted to estimate how important the therapist is. In the early 1990s, Crits-Christoph and colleagues reanalyzed several clinical trials and found that about 8% of the variance in outcomes within treatments was due to the therapists (Crits-Christoph, Baranackie, et al., 1991; Crits-Christoph & Mintz, 1991). Again, 8% may seem small, but is larger than any other single factor; keeping in mind that psychotherapy vis-à-vis no treatment accounts for 13%, the 8% apparently is quite large.
The NIMH treatment of depression collaborative research program (Elkin et al., 1989), the most extensive clinical trial of psychotherapy ever conducted, provides an example of the importance of therapists. The two psychotherapy arms, CBT and IPT, produced similar outcomes. Indeed, zero percent of the variability in outcomes was due to the treatment (CBT vs. IPT). When multilevel models were applied to those data, therapist effects became apparent (Kim, Wampold, & Bolt, 2006): About 8% of the variability in outcomes within each of the psychotherapies was due to the therapist. That is to say, some CBT and some IPT therapists, despite being selected for their expertise, trained to adhere to the respective manuals, and supervised, consistently—across their caseloads—produced better outcomes than others. The therapist variability in that study was similar in magnitude to the estimates made earlier by Crits-Christoph et al. (1991), although the Kim et al. (2006) result was not without some controversy (Elkin, Falconnier, Martinovich, & Mahoney, 2006; Wampold & Bolt, 2007). Additional analyses of clinical trials also have detected sizable therapist effects (e.g., Huppert et al., 2001). Interestingly, in a reanalysis of the NIMH treatment of depression collaborative research program pharmacotherapy arms (i.e., antidepressants vs. pill placebo), effects due to the prescribing psychiatrist became apparent (McKay, Imel, & Wampold, 2006). In those arms, the psychiatrist met with the client weekly for about 30 minutes. The results indicated that the antidepressant was significantly more effective than the pill placebo, accounting for about 3% of the variability in outcomes. However, therapist effects were about 9%—and larger than the antidepressant effect. Indeed, the most effect psychiatrists had better outcomes giving the placebo than did the poorer psychiatrists giving the antidepressant! This is all the more surprising because this effect was produced by limited contact between client and psychiatrist.
It might be expected that therapists in practice would be more variable in their outcomes than therapists in clinical trials, in which the treatment is manualized, would typically be selected for their skills, and would be trained, monitored, and supervised. Estimating therapist effects in a large data set from a managed care context, Wampold and Brown (2005) found that 5% of the variability in outcomes was due to differences among therapists, which is somewhat surprising because this is less than typically found in clinical trials. However, in naturalistic settings, the heterogeneity of the clients makes it more difficult to account for effects. Nevertheless, 5% is clinically important. Wampold and Brown illustrated therapist effects by ranking therapists in the first time period in order of the outcomes they produced with their clients and then examining the therapist outcomes in the second time period. The top quartile of therapists in the first time period, identified with as few as three cases, had much better outcomes than the bottom quartile. The pre- to posttreatment effects sizes were twice as large for the top quartile than for the bottom quartile: Those therapists who attained better outcomes in the first time period had much better outcomes with all types of clients in the second time period. Therapist variability in naturalistic settings also have been found in a number of other studies (Lutz, Leon, Martinovich, Lyons, & Stiles, 2007; Okiishi, Lambert, Nielsen, & Ogles, 2003; Saxon & Barkham, 2012).
It is clear from the clinical trials and the naturalistic studies that some therapists consistently achieve better outcomes than others, a conclusion that has now been established with a comprehensive meta-analysis (Baldwin & Imel, 2013). How the therapist delivers a particular treatment rather than the treatment itself makes a difference. This raises the inevitable question: What are the characteristics and actions of effective therapists? Depressingly, after decades of research, relatively little is known about the characteristics and actions of effective therapists, which has been exacerbated by a decline of research in this area (Beutler et al., 2004). There is little support to suggest that demographics (i.e., age, gender, ethnicity), theoretical orientation, professional training (i.e., experience, degree), or self-reported skills are related to outcomes (Wampold, Baldwin, Holtforth, & Imel, 2017).
Recall that Baldwin et al. (2007) found that more effective therapists are able to form better alliances across a range of clients. Indeed, the differences in average alliances for therapists completely explained the differences in outcomes among the therapists. So, one of the characteristics of effective therapists is that they are able to form a strong alliance across a range of clients, including interpersonally challenging clients.
Anderson, Ogles, Patterson, Lambert, and Vermeersch (2009) used an interesting method to identify the characteristics and actions of effective therapists. Instead of using material from therapy sessions or asking therapists to provide information, they presented videos of challenging clients (i.e., a stimulus that was constant across all therapists) to 25 therapists at a college counseling center, and the therapists recorded their responses to the clients. The responses then were coded for what the authors called facilitative interpersonal skills (FIS), which included verbal fluency, emotional expression, persuasiveness, hopefulness, warmth, empathy, alliance-bond capacity, and problem focus. Those therapists with higher FIS ratings had better outcomes with their clients. This is the first study that has been able to identify characteristics of therapists outside of therapy that would predict actual therapy outcomes. Interestingly, the therapists’ self-reported social skills did not predict their outcomes with clients. It appears that what differentiates more effective from less effective therapists is their interpersonal skills in interpersonally challenging situations. In a prospective study, Anderson, McClintock, Himawan, Song, and Patterson (2016) found that FIS measured in the first few weeks of psychology training predicted the trainees’ outcomes when they saw clients 2 or 3 years later.
The findings of Anderson et al. (2016) were supported by a recent study of therapy trainees. Schöttke, Flückiger, Goldberg, Eversmann, and Lange (2017) assessed psychotherapy trainees using two protocols. In the first, therapy-related interpersonal skills were assessed in a group discussion after watching a disturbing video, which is an interpersonally challenging situation. In the second protocol, the same skills were assessed by a single expert in an interview format. Interpersonal skills assessed with the first protocol (i.e., in the interpersonally challenging environment) predicted the trainees’ outcomes over a 5-year period, even when controlling for many other therapist variables, including theoretical orientation, age, gender, and amount of supervision; and client characteristics, including gender, comorbidity, severity, and personality impairment. However, social skills exhibited in an interview with an expert did not predict outcomes. That study illustrated that, again, effective therapists have a set of sophisticated interpersonal skills that are demonstrated in challenging interpersonal situations.
One of the common factors discussed earlier is that the rationale and treatment should be acceptable to the client. Said another way, the intervention should be compatible with the client’s culture, attitudes, values, and characteristics (Imel & Wampold, 2008; Wampold, 2001b; Wampold, Imel, et al., 2006). Sufficient evidence from clinical trials suggests that many clients drop out of treatment before the end of treatment (Westen & Morrison, 2001) due, to some extent, to the fact that they did not find the treatment agreeable. Initial engagement in the therapeutic process is critical, which has much to do with the client’s preference for treatment (Elkin, Yamaguchi, & Arnkoff, 1999; Iacoviello et al., 2007; Swift, Callahan, & Vollmer, 2011) and expectations about improvement (Connolly Gibbons et al., 2003).
It was noted earlier that matching treatment to the particular psychological deficit has not been supported. However, there is some accumulating evidence that matching treatment to personality, coping styles, and motivation do improve completion of therapy and result in better outcomes. Beutler, Harwood, Michelson, Song, and Holman (2011) reviewed literature that suggested that resistant clients fare better with unstructured treatments, whereas more compliant clients fare better with relatively more structured treatments. In addition, Beutler, Harwood, Kimpara, Verdirame, and Blau (2011) concluded that clients with externalizing disorders benefit more from treatments that focus on skill building and symptom change, whereas those who are self-critical and avoid emotion benefit from relatively more from treatments that focus on interpersonal relations (including with the therapist) and are more insight oriented. Norcross, Krebs, and Prochaska (2011) found that clients who exhibit lack of readiness for change benefit from treatments that focus on motivation and do not pressure clients to take immediate action. Furthermore, treatments that are adapted to the client’s culture, particularly with regard to tailoring the explanation of the treatment to the client’s cultural beliefs, are more effective than nonadapted treatments (Benish, Quintana, & Wampold, 2011; Huey, Tilley, Jones, & Smith, 2014).
One common factor often discussed is the client. The notion is a simple one: It is the client who makes therapy work (Bohart & Tallman, 1999; Bohart & Wade, 2013; Clarkin & Levy, 2004; Duncan, Miller, & Sparks, 2004; Tallman & Bohart, 1999). A distressed client with sufficient motivation and other requisite resources (e.g., adequate social support, economic resources, ego strength), it is conjectured, will use therapy with a relatively skilled clinician to make changes in his or her life, regardless of the treatment offered (Tallman & Bohart, 1999). Indeed, many individuals make significant life changes using self-help sources (i.e., without the help of a healer) with nontherapy persons, such as religious figures (e.g., ministers, rabbis, mullahs) or with various indigenous or alternative healing practices. According to those who emphasize client factors, the therapist and the treatment provide the necessary conditions for clients, who then use the therapist and the treatment to make desired changes.
Psychotherapy is a transactionable endeavor, and it is difficult to parse the various contributions to this process. Common factors, such as the alliance, are created in the interaction with contributions of both the therapist and the client. Determining whether therapeutic success is due to the therapist’s actions or the client’s reactions is extremely difficult (see Baldwin et al., 2007, for an attempt to do so in terms of the alliance). In any event, to be useful, knowledge about how clients make use of therapy is necessary. Several characteristics of clients have been shown to be related to better outcomes in psychotherapy, including greater readiness to change, more psychological resources (i.e., great ego strength), less perfectionism, and higher levels of psychological mindedness (Clarkin & Levy, 2004). In addition, clients with greater initial severity generally will be more distressed at the end of therapy, although they may have made more progress. However, therapists generally do not select their clients based on the presence of characteristics that predict better outcomes because therapists want to assist all clients who present to therapy, and the absence of characteristics that lead to better prognoses often are the factors that form the basis of the client’s problems! Clearly, more research is needed about how clients make use of therapy.
The review of the literature on common factors was necessarily brief because of the multitude of common factors and the different conceptualizations of the common factors. Research exists on such topics as empathy (Elliott, Bohart, Watson, & Greenberg, 2011), positive regard (Farber & Doolin, 2011), expectations (Arnkoff, Glass, & Shapiro, 2002; Constantino, Ametrano, & Greenberg, 2012; Constantino, Glass, Arnkoff, Ametrano, & Smith, 2011; Greenberg, Constantino, & Bruce, 2006) and congruence (Kolden, Klein, Wang, & Austin, 2011), for example. The research on these and other areas is complex because of the difficulties inherent in studying a variable that cannot be manipulated experimentally. For example, empathy is associated with better outcomes, but without disentangling client and therapist contributions to empathy (i.e., Is empathy a therapist-offered condition or is it elicited from the therapist by the client?), understanding how empathy is involved in the process of psychotherapy is problematic.
Despite the difficulties in researching the common factors, there appears to be relatively strong evidence that common factors are important to the process and outcome of psychotherapy (Laska, Gurman, & Wampold, 2014; Wampold & Imel, 2015). To believe otherwise might be surprising because even the most technically minded among us recognize that a client who has a strong bond with his or her therapist and agrees with the goals and tasks of therapy will have a better prognosis than a client who does not. The tension is one of emphasis: Advocates of particular treatments emphasize the effects of specific aspects of the treatment, recognizing that the common factors are necessary, whereas adherents of common factors models emphasize the commonalities, recognizing that a cogent and coherent treatment structure is necessary and that the ingredients are powerful aspects of a treatment.
Although the effectiveness of psychotherapy in the real world is well documented, a significant proportion of clients either do not make adequate progress in therapy or deteriorate. By some accounts, 5% to 10% of clients in therapy deteriorate (Bergin, 1971; Hansen, Lambert, & Forman, 2002; Lambert, 2013). In the past decade, there has been an increased emphasis on using outcomes in routine practice to improve outcomes (Lambert, 2010, Chapter 6; Lambert, Hansen, & Finch, 2001; Lambert, Harmon, Slade, Whipple, & Hawkins, 2005; S. D. Miller, Duncan, & Hubble, 2005). Of particular interest is the research on providing feedback to therapists about the outcomes achieved by their clients. In a series of studies, Lambert and colleagues (Harmon et al., 2007; Lambert, Harmon, et al., 2005; Lambert, Whipple, et al., 2001; Lambert, Whipple, et al., 2002; Whipple et al., 2003) investigated the effects of informing therapists whether a particular client is making expected progress (i.e., improving at a rate expected for clients of a given level of severity), is not making expected progress but is improving, or is deteriorating. That research has shown that the clients of therapists who receive this simple feedback attain better outcomes at the end of treatment than clients of therapists who do not receive feedback. This effect is particularly large for clients who are not making expected progress (Lambert & Shimokawa, 2011). However, it appears that receiving such feedback does not help therapists to become more effective over time (Goldberg, Rousmaniere, et al., 2016; Tracey, Wampold, Lichtenberg, & Goodyear, 2014). Indeed, it is difficult to become an expert therapist (Tracey et al., 2014).
The previously discussed two studies on therapist interpersonal skills (i.e., Anderson et al., 2016; Schöttke et al., 2017) suggested that such skills demonstrated early in one’s training determine how good a therapist a trainee will be years later. Adding to this finding is evidence that therapists do not improve with experience (i.e., with time or number of cases), even when they receive feedback on client progress (Erekson, Janis, Bailey, Cattani, & Pedersen, 2017; Goldberg, Rousmaniere, et al., 2016). However discouraging this may be, there also is evidence that therapists can improve; the key seems to be related to practice. Experts in many fields, including sports, music, and chess, are characterized by practice (Ericsson, Krampe, & Tesch-Römer, 1993; Ericsson & Lehmann, 1996). Deliberate practice involves having a coach, practicing particular skills with feedback, and making gradual improvement. Therapists who express professional self-doubt by questioning their competence and have a desire to be better have better outcomes with their clients (Nissen-Lie, Monsen, Ulleberg, & Rønnestad, 2013; Nissen-Lie, Rønnestad, et al., 2017). Evidence indicates that those therapists who spend time outside of therapy attempting to improve by seeking supervision and consultation, reviewing videos of their work, reading books on therapy and attending workshops have better outcomes (Chow et al., 2015). Moreover, agencies that incorporate deliberate practice into their work gradually achieve better outcomes (Goldberg, Babins-Wagner, et al., 2016). It is becoming clear that to become a better therapist, one must continue to practice (Rousmaniere, Goodyear, Miller, & Wampold, 2017).
The research evidence has established, without much debate, that psychotherapy is effective. However, with some exceptions, efforts to identify the importance of particular specific ingredients of manualized treatments have not been successful. On the other hand, there is increasing evidence that the alliance, interpersonal skill, and the person of the therapist are important factors in psychotherapy. Clearly, additional research is needed to understand how psychotherapy works. A critical issue is determining how psychotherapy delivered in the real world can be improved both in terms of outcomes of clients in psychotherapy and delivery of psychotherapy to those in need of treatment, including culturally diverse groups. It appears that deliberate practice with feedback about performance would be beneficial (Chow et al., 2015; Goldberg, Babins-Wagner, et al., 2016; Rousmaniere et al., 2017).