Introduction
Decades of empirical research demonstrate a positive relation between common factors and treatment response (Wampold, Reference Wampold2015). For example, both the working alliance (mutual agreement on goals, tasks and bond; Bordin, Reference Bordin1979) and outcome expectancy (the extent to which a person believes treatment will be successful) are robust predictors of symptom reduction across a wide range of interventions and disorders (for recent meta-analyses, see Constantino et al., Reference Constantino, Arnkoff, Glass, Ametrano and Smith2011; Flückiger et al., Reference Flückiger, Del Re, Wampold and Horvath2018). Very little is known, however, about the mechanisms through which common factors effect change (Kazdin, Reference Kazdin2005), which is needed so that treatments may best capitalize on the therapeutic benefits of common factors (Oddli et al., Reference Oddli, Nissen-Lie and Halvorsen2016).
One example of such work is research examining the working alliance as a mechanism through which outcome expectancy leads to symptom improvement. The quality of the working alliance has been shown to mediate the relation between outcome expectancy and symptom improvement across a range of psychological concerns, including depression, anxiety, complicated grief, personality disorders, and interpersonal problems (Abouguendia et al., Reference Abouguendia, Joyce, Piper and Ogrodniczuk2004; Johansson et al., Reference Johansson, Høglend and Hersoug2011; Joyce et al., Reference Joyce, Ogrodniczuk, Piper and McCallum2003; Meyer et al., Reference Meyer, Pilkonis, Krupnick, Egan, Simmens and Sotsky2002; Sauer-Zavala et al., Reference Sauer-Zavala, Boswell, Bentley, Thompson-Hollands, Farchione and Barlow2018; Vȋslă et al., Reference Vȋslă, Constantino, Newkirk, Ogrodniczuk and Söchting2018), but has yet to be tested for those receiving cognitive behavioural therapy for social anxiety disorder. Social anxiety disorder is an interesting disorder in which to examine the potential mediating role of the working alliance, because negative evaluation is the core fear of the disorder, and psychotherapy is an interpersonal process in which this core fear may be triggered (Hayes et al., Reference Hayes, Hope, VanDyke and Heimberg2007). By its nature, treatment for social anxiety disorder may represent a distinct context in which the working alliance is unrelated to therapeutic change. For example, social anxiety is associated with increased fear of expressing strong emotions, increased desire to avoid conflict, increased conflict avoidance, decreased assertiveness, increased over-reliance on others, and increased fear of rejection (Davila and Beck, Reference Davila and Beck2002). Social anxiety has also been shown to have a negative impact on interpersonal communication and interaction, including reduced conversational interruptions (Natale et al., Reference Natale, Entin and Jaffe1979), generation and expression of fewer ideas (Camacho and Paulus, Reference Camacho and Paulus1995), more variability in interpersonal behaviours (Rappaport et al., Reference Rappaport, Moskowitz and D’Antono2014), and limited vulnerability and intimacy in conversation (Alden and Taylor, Reference Alden and Taylor2004). As such, the working alliance may not represent an effective mechanism of therapeutic change among individuals with clinically significant symptoms of social anxiety.
The purpose of this study is to test the hypothesis that the working alliance mediates the relation between outcome expectancy and symptom improvement in a sample of individuals with a primary diagnosis of social anxiety disorder, while controlling for therapist effects. Therapist effects occur when some therapists’ clients report achieving consistently better outcomes than others (Wampold, Reference Wampold2015). Notably, the working alliance is a distinct construct from therapist effects. This study is the first to examine a potential mechanism through which common factors exert influence in treatment for social anxiety disorder.
Method
Participants
Participants were 54 individuals diagnosed with social anxiety disorder. About half of the participants (n = 32) met criteria for the generalized subtype. Eligibility criteria are reported in the parent study publication (Anderson et al., Reference Anderson, Price, Edwards, Obasaju, Schmertz, Zimand and Calamaras2013). Power was assessed based on recommendations from Fritz and MacKinnon (Reference Fritz and MacKinnon2007) and indicated a sample size of 59 is needed to detect a significant indirect effect. Predicted effect sizes were based on previous findings in the literature (Flückiger et al., Reference Flückiger, Del Re, Wampold and Horvath2018; Price and Anderson, Reference Price and Anderson2012).
Measures
Diagnoses were determined via administration of the Structured Clinical Interview for the DSM-IV (SCID; First et al., Reference First, Spitzer, Gibbon and Williams1994). Outcome expectancy was assessed with an adaptation of the four-item Credibility/Expectancy Questionnaire (CEQ; Borkovec and Nau, Reference Borkovec and Nau1972). Quality of the working alliance was assessed using the Working Alliance Inventory-Short Form (WAI-SF; Tracey and Kokotovic, Reference Tracey and Kokotovic1989). Three measures were used to assess symptoms of social anxiety: the Personal Report of Confidence as a Speaker (PRCS; Paul, Reference Paul1966), the Fear of Negative Evaluation-Brief Form (FNE-B; Leary, Reference Leary1983), and the self-report version of the Liebowitz Social Anxiety Scale (LSAS-SR; Liebowitz, Reference Liebowitz1987). Internal consistencies in the present study ranged from α = .61 to α = .96 across all measures and time points. All measures in the study are widely used and have demonstrated good psychometric properties.
Design
The data used in the present study come from a randomized clinical trial comparing virtual reality exposure (VRE) therapy and exposure group therapy (EGT) for social anxiety disorder (Anderson et al., Reference Anderson, Price, Edwards, Obasaju, Schmertz, Zimand and Calamaras2013). Ethical approval for this study was granted by the Georgia State University Institutional Review Board (reference no. 000000716). Mediation analyses were conducted using Andrew Hayes’ Process Macro (Hayes, Reference Hayes2013). Outcome expectancy served as the independent variable, working alliance as the mediating variable, and three measures of social anxiety symptoms as the outcome variables. Dummy codes were used to control for the possibility of therapist effects. Pre-treatment symptoms were controlled for in all analyses. Missing data were handled with last observation carried forward (LOCF), which has proven a more conservative estimate of treatment effects than other imputation methods (Unnebrink and Windeler, Reference Unnebrink and Windeler2001).
Procedure
Eligible participants were randomly assigned to one of three groups: virtual reality exposure therapy, exposure group therapy, or a wait-list control [see Anderson et al. (Reference Anderson, Price, Edwards, Obasaju, Schmertz, Zimand and Calamaras2013) for participant flow]. The sample used in the present study includes wait-list participants who were re-randomized and completed one of the two active treatments. Outcome expectancy was assessed during the first session following presentation of the treatment rationale. The working alliance was measured at every treatment session. Symptom measures were assessed at pre- and post-treatment.
Treatments
There were five study therapists: two licensed psychologists and three doctoral students in clinical psychology. One therapist was male and the other four were female. Three therapists were from the USA and were White, one therapist was from the USA and was Black, and one therapist was from South Africa and was White. Each therapist conducted both treatments. All treatment sessions were video recorded, and a randomly selected subset of treatment sessions (14%) were reviewed by the treatment developers for fidelity and competence, which was high. Compliance was good for each treatment, with 92 and 93% of the essential elements of the protocol completed for VRE and EGT, respectively. There were seven treatment groups ranging from two to four members. For individual treatment in the virtual reality condition, therapists saw an average of eight participants over the course of the study. No significant differences in working alliance (measured at any session) were detected based on the therapist or by group membership.
Treatments were designed to be as similar as possible except for the delivery of exposure (virtual reality vs exposure group therapy). Both treatments included a treatment rationale, psychoeducation and cognitive therapy [see Anderson et al. (Reference Anderson, Price, Edwards, Obasaju, Schmertz, Zimand and Calamaras2013) for more details], and consisted of eight sessions.
Results
Descriptive statistics are shown in Table 1. No significant differences were detected between those who completed an active treatment and those who dropped out from an active treatment in outcome expectancy or working alliance when measured at session 1.
Table 1. Means and standard deviations for all measures used in mediational analyses
FNE-B, Fear of Negative Evaluation-Brief Form; LSAS, Liebowitz Social Anxiety Scale.
Bias corrected bootstrap confidence intervals were used to test the hypothesis that the quality of the working alliance mediates the relation between outcome expectancy and symptom improvement (Hayes, Reference Hayes2013). Bootstrapping approaches place greater emphasis on the indirect effect, are less susceptible to violations of the normality assumption, and are significantly less prone to Type II error relative to more traditional approaches to mediation such as the Baron and Kenny (Reference Baron and Kenny1986) approach (Hayes, Reference Hayes2013).
The indirect effect of the working alliance on symptom improvement was not significant across all measurements of the working alliance (sessions 1 to 8) and across all symptom measures (FNE-B, PRCS, LSAS-SR).
Analysis of direct effects showed that working alliance (measured at any time point) was not associated with any measure of symptom improvement. This was surprising given the robust relation between the working alliance and outcomes reported in the literature (Flückiger et al., Reference Flückiger, Del Re, Wampold and Horvath2018). Notably, null findings do not mean that an effect is absent and cannot inform confidence in a theory. In contrast, estimation of Bayes can be used to evaluate how strongly the data support the null or alternative hypothesis (Dienes, Reference Dienes2014). A Bayes factor of three or larger reflects high confidence in the alternative hypothesis, and a Bayes factor of 0.33 or less reflects high confidence in the null hypothesis. Any value between 0.33 and 3.00 reflects insensitive data. We estimated Bayes factors for the relation between the working alliance and symptom reduction, while controlling for the effect of outcome expectancy and therapist effects. Bayes factors were all between 0.33 and 3.00, reflecting insensitivity in the data (FNEB: 0.61–1.9; PRCS: 0.49–1.29; LSAS: 0.59–2.73).
Discussion
The data in the present study were consistent in producing null results – across the types of exposure-based treatment (exposure group therapy, virtual reality exposure therapy), across all treatment sessions, and for three different measures of symptom improvement. The general lack of an indirect effect of working alliance on the relation between outcome expectancy and symptom improvement is surprising because it inconsistent with the handful of empirical studies examining this question (Abouguendia et al., Reference Abouguendia, Joyce, Piper and Ogrodniczuk2004; Johansson et al., Reference Johansson, Høglend and Hersoug2011; Joyce et al., Reference Joyce, Ogrodniczuk, Piper and McCallum2003; Meyer et al., Reference Meyer, Pilkonis, Krupnick, Egan, Simmens and Sotsky2002; Sauer-Zavala et al., Reference Sauer-Zavala, Boswell, Bentley, Thompson-Hollands, Farchione and Barlow2018; Vȋslă et al., Reference Vȋslă, Constantino, Newkirk, Ogrodniczuk and Söchting2018). These studies do not, however, explore this mediational model among a sample of individuals with a primary diagnosis of social anxiety disorder. Treatment for social anxiety disorder may represent a unique context in which the working alliance is less relevant for, or unrelated to, symptom change. This would underscore Kazdin’s (Reference Kazdin2005) assertion that the impact of common factors may vary across treatment contexts.
Although we did not specifically hypothesize a positive direct relation between the quality of the working alliance and symptom improvement, it is unexpected that there was none given meta-analytic research, which concludes there is a robust relation between the working alliance and treatment outcome (Flückiger et al., Reference Flückiger, Del Re, Wampold and Horvath2018). However, all other known studies of this relation for cognitive behavioural treatments for social anxiety disorder (Andersson et al., Reference Andersson, Paxling, Wiwe, Vernmark, Felix, Lundborg and Carlbring2012; Mörtberg, Reference Mörtberg2014; Woody and Adessky, Reference Woody and Adessky2003) have, like the current study, produced null results. Similarly, a meta-analysis examining the effect of exposure treatment on symptoms of anxiety reported null results from a moderator analysis of the working alliance, although only two studies were included in this analysis (Parker et al., Reference Parker, Waller, Gonzalez Salas Duhne and Dawson2018).
It is important to note, however, that a null result, or even a collection of studies reporting null results, is not a guarantee that an effect is absent. Our analyses of Bayes factors for the relation between the working alliance and symptom reduction, while controlling for outcome expectancy and therapist effects, were inconclusive. Our data were insensitive and unable to provide sufficient evidence to distinguish between the null and alternative hypothesis. These findings highlight the need for further evaluation of the impact of the working alliance on treatment for social anxiety disorder.
The current study has several limitations. First, all variables of interest were operationalized by self-report measures collected from a single source: the participant. In addition, although this study’s measure of outcome expectancy has been widely used, the measure may be more of an indicator of treatment credibility, a related, but distinct construct (Devilly and Borkovec, Reference Devilly and Borkovec2000). The sample size (n = 54) reflects limited power. Critically, neither outcome expectancy nor working alliance were manipulated in the present study, which means that it is inappropriate to draw conclusions regarding causal relations. Finally, another notable limitation is the treatment formats used in the present study. Working alliance theorists emphasize the role of traditional dyadic format relations (e.g. Muran and Safran, Reference Muran, Safran and Magnavita2002). Other studies reporting on the relation between the working alliance and symptom improvement in treatment for social anxiety disorder also utilize this treatment approach (Mörtberg, Reference Mörtberg2014; Woody and Adessky, Reference Woody and Adessky2003). Mörtberg (Reference Mörtberg2014) reported a null relation between the working alliance in both group and individual therapy formats. Woody and Adessky (Reference Woody and Adessky2003) evaluated the working alliance and group cohesion, both of which were unrelated to outcome.
Despite these limitations, the current study is one among very few to investigate the mechanisms through which common factors influence treatment response, and the first to do so in a sample of adults receiving treatment for social anxiety disorder. The results emphasize the need for further exploration of the mechanisms through which common factors relate to treatment response, particularly in unique contexts that may influence the relevance of common factors such as treatment for social anxiety disorder.
Acknowledgements
None.
Financial support
This study was funded by the National Institute of Health (R42 MH 60506-02), awarded to the second author. The first author has received funding from the Brains and Behavior Fellowship.
Conflicts of interest
Amanda Draheim and Page Anderson have no conflicts of interest with respect to this publication.
Ethical statements
All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study.
Key practice points
(1) Working alliance is considered a common mechanism across all treatment contexts.
(2) Current findings do not support working alliance as a mechanism of treatment for social anxiety disorder.
Comments
No Comments have been published for this article.