The sister concepts of working alliance and real relationship: A meta-analysis

ABSTRACT

The real relationship and working alliance are considered sister concepts, because, even though they are distinct elements of the therapeutic relationship, they seem to be highly interrelated, especially regarding the bond element of the relationship. Both concepts are considered to contribute significantly to the therapy outcomes, but some questions remain regarding their relationship, similarities, and conceptual differences. The present study aimed to examine the association of real relationship and working alliance across studies and analyze their contribution to treatment success. For this purpose, we conducted a meta-analysis on these two constructs and performed a quantitative analysis to differentiate the contribution of real relationship and working alliance for psychotherapy outcomes. Results from the meta-analysis confirmed the theorized association between real relationship and working alliance, revealing an overall correlation of r = .66. Based on our quantitative analysis, real relationship seems to be as good a predictor as the working alliance, although both benefit from each other. The research findings underline the importance of considering the contribution of these two concepts for therapeutic outcomes. Therapists should pay attention to both real relationship and the working alliance as key factors that improve therapy success.

KEYWORDS:

• Therapeutic relationship
• real relationship
• working alliance
• therapy outcomes
• common factors

Introduction

Throughout the years, some therapists have considered that the relationship developed with their clients is the essence of effective treatment, while others argue that a good therapist-patient relationship provides significant leverage for the implementation of therapy techniques (Gelso & Carter, 1994). Whether the relationship is the essential ingredient of therapy or a means to an end, it is well established that it plays an important role in therapy (Norcross & Lambert, 2019). Norcross (2011) states that the therapeutic relationship makes substantial and consistent contributions to clients’ success across different psychotherapeutic orientations (e.g. cognitive, psychodynamic, humanistic), and it accounts for why clients improve – or fail to improve – more than the techniques of each treatment method (Wampold & Imel, 2015).

According to Gelso (2014), all psychotherapy relationships consist of three interlocking elements: a real relationship, a working alliance, and a transference configuration (both transference and countertransference) – referred to as the tripartite model. These elements were rooted in psychoanalytic theory, but Gelso argues that each of these are present across all theoretical orientations and exists from the first moment of contact between therapist and patient. These three relationship components are independent, but they do not operate independently; they interact constantly and, to some extent, overlap throughout therapy (Gelso & Carter, 1994). In the present study, we focus only on the concepts of real relationship and working alliance because the interrelation between the two concepts, and even whether they are two independent concepts, has proved controversial (Gelso, 2011, 2019). We begin by examining conceptualizations of the two constructs and their relationship to treatment outcome.

Conceptualizations of the real relationship and the working alliance: similarities and differences

Real relationship may be defined as “the personal relationship existing between two or more people as reflected in the degree to which each is genuine with the other and perceives the other in ways that befit the other” (Gelso, 2009, pp. 254–255). Two components emerge from this definition: genuineness and realism. Genuineness defines the psychotherapy participants’ authenticity with each other or the extent to which they are truly themselves (Gelso et al., 2012). According to Gelso (2009), the concept of genuineness pertains to both participants, with their personal characteristics, but also to the quality of the relationship itself. On the other hand, realism is the experiencing or perceiving of the other in ways that befit him or her, rather than projections of the individual based on his/her fears or wishes (Gelso, 2009). Gelso (2009) separates realism and genuineness for theoretical and measurement reasons but acknowledges that they must be closely intertwined. Perceiving and experiencing the other as he/she truly is, requires him/her to be genuine; likewise, how genuine one is will be strongly influenced by the extent to which one feels understood accurately by the other.

To better understand the real relationship, we also need to consider two other concepts: magnitude and valence. Magnitude refers to the amount of real relationship (how much genuineness and realism) which exists, both overall and on a moment-to-moment basis (Gelso, Kivlighan, & Markin, 2018). Valence concerns how positive or negative the participants’ feelings and thoughts are towards one another. Each participant may experience and perceive the other positively or negatively in terms of realism and genuineness.

Realism and genuineness are the key elements of the real relationship, and this conceptualization has been crucial for the current scientific knowledge and empirical research (Gelso, Kivlighan, & Markin, 2018). The combination of genuineness and realism, including their magnitude and valence, results in the strength of the real relationship. The more positive the therapist’s and the client’s genuine and realistic feelings for one another are, the stronger is the real relationship (Gelso & Kline, 2019).

Along with the real relationship, alliance is a common relational factor that has been widely studied. The term alliance can be preceded by therapeutic, working, or helping (Flückiger, Del Re, Wampold, & Horvath, 2018). The concept dates back to Freud and his recognition of the importance of the client’s conscious attachment to the person of the therapist (Flückiger, Del Re, Wampold, & Horvath, 2018).

Bordin (1979) proposed a pantheoretical version of the alliance called “working alliance”, which is currently the most consensually agreed upon. He defined working alliance as a concept originated from psychoanalytic theories that describes the relationship quality as a result of the joint collaboration between client and therapist at a conscious level. For Bordin, the working alliance is a collaboration between therapist and client that rests on three components: agreement on therapeutic goals (the goals each client needs and/or wants to achieve in therapy), consensus on the tasks that make up therapy (needed to obtain the established goals), and a bond between the dyad (the emotional connection dimension) (Bordin, 1979). In his tripartite model, Gelso (2009, 2014) also embraced Bordin’s definition of working alliance because the focus is on the working aspect of alliance – allowing it to be differentiated from other components which are not directly linked to a working collaboration.

Since 2002, alliance has been considered as an empirically validated common factor by the American Psychological Association (APA) guidelines (Norcross, 2002). Research has consistently found a positive relation between alliance and treatment outcomes, with its effect size being similarly significant for different therapeutic models, such as psychodynamic, interpersonal, cognitive behavioral and eclectic therapies (Flückiger, Del Re, Wampold, & Horvath, 2019). In fact, several authors and therapeutic models consider alliance as not only a common factor but a determining component for psychotherapeutic success (e.g. Flückiger, Del Re, Wampold, & Horvath, 2018; Leichsenring, Steinert, & Crits-Christoph, 2018). Furthermore, studies on the alliance have proven its positive impact in different age groups, such as adults (Flückiger, Del Re, Wampold, & Horvath, 2019), children, and adolescents (Karver, De Nadai, Monahan, & Shirk, 2019), as well as in different settings, such as individual therapy (Flückiger, Del Re, Wampold, & Horvath, 2019) and couple and family therapies (Friedlander, Escudero, Welmers-van de Poll, & Heatherington, 2019).

Regarding the measurement of alliance, there are approximately 30 different instruments that aim to evaluate this construct. The most widely applied are the Helping Alliance Questionnaire (HAQ; Alexander & Luborsky, 1986), the Vanderbilt Psychotherapy Process Scale (VPPS; Suh, Strupp, & O’Malley, 1986), the California Psychotherapy Alliance Scale (CALPAS; Marmar, Horowitz, Weiss, & Marziali, 1986), and the Working Alliance Inventory (WAI; Horvath & Greenberg, 1989). However, about 69% of the most recent scientific investigations use the WAI and its shortened versions, since it is the most reliable measure of the construct, and its items are consistent with Bordin’s (1979) alliance definition.

Despite of being theoretically distinct constructs with different contributions to outcomes (e.g. Bhatia & Gelso, 2018), working alliance and real relationship have been seen as strongly interrelated (Gelso & Kline, 2019; Gelso, Kivlighan, & Markin, 2018). The psychoanalyst Greenson (1967) was the first to analyse these two concepts in interrelation. According to him, the working alliance arises from the real relationship, since real relationship is a universal aspect of all human relationships and the working alliance – considered as an artifact of psychotherapy – only exists to allow the therapy work to be done (Greenson, 1967). Later, Gelso and Carter (1994) proposed a different conclusion and stated that they appear at same time and influence each other over the course of treatment.

Thus, while the alliance integrates aspects of collaborative work, the real relationship focuses especially on the personal connection (Gelso, 2009, 2014), with the bond being the common element between them. Gelso and Kline (2019) argue that we could consider the bond element – present in both working alliance and real relationship – as a working bond and a personal bond, respectively. The working bond refers to the collaborative effort between therapist and client to make therapeutic work successful. It implies the client’s trust in the therapist’s capacity to understand him/her and his/her ability to help and the therapist’s motivation to conduct the work effectively. Personal bond, on the other hand, is a connection that is not directly linked to therapeutic work, but to the interpersonal connection between two people who spend time together and appreciate each other. In this way, bond is the key feature that makes the two concepts emerge simultaneously and work together – the patient is inclined to be motivated to do the work of therapy when he/she personally connects to the therapist and the two participants work well together to create a sense of personal relationship (Gelso & Kline, 2019). Due to the theoretical distinction but simultaneous overlap (through bond), they have been called “sister concepts” (Gelso & Kline, 2019; Gelso, 2014; Gelso, Kivlighan, & Markin, 2018).

At the same time, there might be times when the two constructs do not work in tandem. Not all therapists feel strongly connected to their clients as persons. The real relationship might be weak at the beginning of treatment, while the working alliance may be solid. It can also happen that the real relationship never becomes strong, but the work can be successful because the working alliance is strong enough. Nonetheless, Gelso and Kline (2019) believe that the results might not be as successful as they could have been if both real relationship and working alliance were strong.

What about a weaker working alliance compared to a strong real relationship? Gelso and Kline (2019) also considered this and concluded that it might be hard for a treatment to be successful if the working alliance is weak, because it concerns the therapy goals, the tasks necessary to attain those goals, and the bond directly related to the therapy work.

Thus, it seems that both the alliance and the real relationship are important factors for psychotherapeutic success. The Third Interdivisional American Psychological Association (APA) Task Force on Evidence-Based Relationships and Responsiveness sought to explore the effective elements of the therapeutic relationship. In doing so, Norcross and Lambert (2019) included chapters that contained meta-analyses about the working alliance and the real relationship (among others) as related to treatment outcome.

Research on the relationship of alliance and real relationship to each other and treatment outcome

Working alliance has been submitted to several meta-analyses to understand its relation to treatment outcome. Horvath and Symonds (1991) conducted the first, revealing an overall alliance-outcome correlation of r = .26. That effect size has proven robust across psychotherapies and decades of research. The following meta-analyses’ correlations varied only slightly over the years: Horvath and Bedi (2002): r = .21, k = 100; Horvath, Del Re, Flückiger, and Symonds (2011): r = .28, k = 190; Martin, Garske, and Davis (2000): r = .22, k = 79. The most recent meta-analysis was conducted by Flückiger, Del Re, Wampold, and Horvath (2018) and revealed a correlation of r = .278, identical to the one conducted by the same authors in 2011, where it was found that the alliance-outcome relationship accounts for about 8% of the variance of treatment outcomes. Alliance has been associated with several different outcomes, such as a decrease in overall symptomatology, global outcome measures, risk behaviors, interpersonal problems, lower dropout and improved psychological well-being. These data are consistent across client diagnosis, although the alliance-outcome association is smaller in the substance use disorder population and in individuals with eating disorders (Flückiger, Del Re, Wampold, & Horvath, 2019).

Regarding the real relationship, Eugster and Wampold (1996) were the first to conduct an empirical study on this concept. In their study, they used a specially devised 8-item measure of the real relationship. Results showed that both therapist and patient ratings of real relationship correlated moderately positively with session evaluation (r ranging from .28 to .64). This first study demonstrated that the real relationship was a promising variable in terms of its potential influence on treatment outcome, but there was still the lack of a reliable, valid, and economical measure to continue the investigation of this concept. In 2005, Gelso and his collaborators developed the first validated instrument that allowed therapists to rate the real relationship: the Real Relationship Inventory – Therapist Form (RRI-T; Gelso et al., 2005). Later, Kelley, Gelso, Fuertes, Marmarosh, and Lanier (2010) developed the client version – the Real Relationship Inventory – Client Form (RRI-C).

The first meta-analytic review of the association between the real relationship and psychotherapy outcomes was conducted by Gelso, Kivlighan, and Markin (2018), including a total of 16 studies. Meta-analysis results revealed a moderate real relationship-outcome association (r = .38, 95% CI [.30, .44], p < .001), which is a larger magnitude than the working alliance-outcome association (small, r = .28, 95% CI [.26, .30], p < .0001) found in the most recent meta-analysis (Flückiger, Del Re, Wampold, & Horvath, 2018). However, this result should be interpreted with caution, given that there are far fewer studies on the real relationship and outcomes than on the working alliance and outcomes. Therefore, the APA task force considered alliance as a demonstrably effective factor for therapeutic change, while the real relationship is still only considered as probably effective (Norcross & Lambert, 2018).

Regarding these two common factors, a meta-analysis that focused on the association between real relationship and working alliance was also conducted, using a subset of nine studies (Gelso, Kivlighan, & Markin, 2019). The nine studies reported the correlation between real relationship and working alliance, and the meta-analysis based on the reported correlations found a significant omnibus effect (r = .58, 95% CI [.51, .64], p < .001). This result supports Gelso’s characterization of the constructs as sister concepts – medium to large correlation, but not identical constructs.

As stated above, from the original sixteen studies that revealed a correlation between real relationship and outcomes, nine also showed a correlation between real relationship and working alliance, and these were used to perform a meta-analysis between the two concepts. However, this meta-analysis, which is related to our research question, was incomplete because it did not include studies of the association between real relationship and working alliance where outcome was not measured.

Objective

The main aims of our study were twofold. First, we sought to clarify the conceptual overlap between real relationship and working alliance. To that end, we replicated the meta-analysis on the association between these two constructs performed by Gelso, Kivlighan, and Markin (2019), with the addition of studies that did not examine outcome. Second, we intended to explore and compare real relationship and working alliance’s contribution to treatment outcomes. In order to accomplish this second aim, we conducted an overview of individual quantitative studies, and then we compared the correlation between real relationship and outcomes with the correlation of working alliance with the outcomes.

Method

Search strategy

We searched the databases Web of Science and PsycINFO until the end of August 2020 using the following terms: “real relationship” and “working alliance”, “therapeutic alliance” or “helping alliance” (Flückiger, Del Re, Wampold, & Horvath, 2018).

The inclusion criteria were: (a) peer-reviewed articles; (b) written in the English language; (c) adult samples; (d) studies of individual therapy; (e) in-person or online therapy; (f) both working alliance and real relationship were measured in the study at least one time. All instruments that assessed the real relationship and the working alliance were included, with no specific measure considered as a criterion for inclusion. The exclusion criteria included papers where (a) the therapeutic relationship was not studied, (b) qualitative studies and (c) case studies, dissertations, and book chapters.

Results

Study selection

The selection process of studies for meta-analysis was made with PRISMA guidelines (Moher et al., 2009), whose diagram can be found in Figure 1.

Figure 1. PRISMA flow diagram of the study selection process.

Data from the initial search resulted in 110 citations, leaving 86 when duplicates were removed. The articles were screened at abstract level, and the remaining 62 articles were screened at full text level against inclusion criteria, leaving us with 23 included articles. This process of screening was done by the first author and audited by the last author. In case of doubts or disagreements, the two debated until a consensus was reached. Table 1 presents the characteristics of all 23 studies.

Table 1. Studies characteristics and key findings.

Author, year; country	Design	Objectives	Participants	N	Key measures	Key findings
Gelso et al. (2005); USA	Cross-sectional	Develop and validate the RRI-T.	Therapists	210	WAI-S; SEQ – Depth andSmoothness	The RRI – T total score and one or both subscale scores correlated significantly with the WAI, the SEQ Depth and Smoothness subscales, emotional insight, intellectual insight, and negative transference.
Fuertes et al. (2007); USA	Cross-sectional	Examine therapist and client ratings of the RR in relation to their WA ratings; Examine the role of therapist and client attachment in the formation of the RR.	Therapists and clients	118	WAI-S; RRI; COM	For both therapists and clients, there was a positive relationship between their ratings of the RR and the WA. With respect to progress, client and therapist ratings of the RR, and not their ratings of the WA, were predictive of their ratings of client progress.
Marmarosh et al. (2009); USA	Longitudinal	Understand how the RR relates to important process and outcome variables (attachment, WA, transference, and treatment outcome) from both the clients’ and therapists’ perspectives.	Therapists and clients	52	WAI-S; RRI; SCL-90-R	Clients’ perceptions of the RR correlated with both therapist and client ratings of WA. Therapist ratings of the RR correlated with only therapist ratings of WA. Therapists’ perceptions of the RR did not correlate with clients’ perceptions of the alliance at the third session of therapy. Therapist-rated RR was the only significant predictor of outcome.
Kelley, Gelso, Fuertes, Marmarosh, and Lanier (2010); USA	Cross-sectional	Develop and validate the RRI-C.	Clients	187	WAI-S	The RRI-C was positively correlated with a measure of clients’ observing ego functions and WA; it was negatively correlated with a scale that measures clients’ tendency to hide their true feelings and to change their behavior to fit in or meet others’ expectations.
Owen, Tao, Leach, and Rodolfa (2011); USA	Cross-sectional	Examine whether clients’ perceptions of their psychotherapists’ multicultural orientation (MCO) were associated with their psychological functioning, WA, and RR scores.	Clients	176	WAI-SR; RRI; SOS-10	Clients’ perceptions of their psychotherapists’ MCO were positively related to the WA and WA significantly mediated the relationship between clients’ perceptions of their psychotherapists’ MCO and client psychological well-being. Clients’ perceptions of the RR was not a significant mediator for the association between clients’ ratings of their psychotherapists’ MCO and psychological well-being, but a strong and positive association was found between clients’ perceptions of their psychotherapists’ MCO and the RR.
Lo Coco, Gullo, Prestano, and Gelso (2011); Italy	Longitudinal	Association of the client- and therapist-rated strength of the RR to the outcome of brief psychotherapy; Extent to which the RR predicts outcome above and beyond the predictive power of the WA.	Therapists and clients	54	WAI-S; RRI; OQ-45	From the clients’ perspective, both the Genuineness of the RR and the Bond scale of the WA were found to relate significantly to treatment outcome when these variables were measured early in treatment. However, neither the therapist-rated RR nor the therapist-rated WA, when measured early in treatment, were significantly associated with outcome.
Hill et al. (2015); USA	Longitudinal	Investigate changes over 12 to 42 months in 23 predoctoral trainees during their externship training in a psychodynamic/interpersonal psychotherapy clinic.	Therapists and clients	191	WAI-SR; RRI; OQ-45.2	Over their time in the clinic, trainees were able to form stronger WA (as rated by both clients and therapists) and stronger RR (as rated by clients).
Kelley (2015); USA	Cross-sectional	Investigate the role of therapy practices and the therapy relationship on lesbian and gay clients’ feelings about their current therapist.	Clients	116	WAI-SR; RRI	Clients’ perceptions of their therapists’ therapy practices, the RR, and the WA were significantly positively related to these clients’ feelings about their therapists. The RR and the WA were strongly correlated.
Kivlighan, Hill, Gelso, and Baumann (2016); USA	Longitudinal	Examine the dyadic associations of clients and therapists in their evaluations of WA, RR, session quality, and client improvement over time in ongoing psychodynamic or interpersonal psychotherapy.	Therapists and clients	97	WAI-SR; RRI; OQ-45	Clients’ and therapists’ ratings of the WA and the RR were significantly and uniquely related to their own ratings of session quality. Client-rated WA and RR were associated with therapist-rated session quality.
Doran, Safran, and Muran (2016); USA	Cross-sectional	Investigate the utility and psychometric properties of the Alliance Negotiation Scale (ANS).	Clients	212	WAI; RRI	The relationship between the ANS and WAI was positive and statistically significant; the ANS and RRI were significantly positively correlated. The correlations point to a substantial overlap between the three measures.
Shafran, Kivlighan, Gelso, Bhatia, and Hill (2017); USA	Longitudinal	Examine the relationship between amount of therapist immediacy in sessions and client post-session ratings of WA, RRI, and session quality, using hierarchical linear modeling (HLM).	Therapists and clients	25	WAI-SR; RRI	More immediacy in a session was related to higher client ratings of session quality for that session. Whereas more immediacy in a session was related to lower client ratings of WA early in treatment, more immediacy in a session was related to higher client ratings of WA later in treatment. No relationship was found between immediacy and RR.
Kivlighan, Kline, Gelso, and Hill (2017); USA	Longitudinal	Examine how congruence and discrepancy in clients’ and therapists’ ratings of the WA and RR were related to client-rated session quality.	Therapists and clients	167	WAI-SR; RRI; SES	RR – WA discrepancy is a common occurrence at the session level. More than 50% of sessions had discrepant client RR – WA scores, and almost 45% of the sessions had discrepant therapist ratings.
Bhatia and Gelso (2017); USA	Cross-sectional	Examine how 3 elements of the therapy relationship (WA, RR, and transference) during the termination phase relate to perceived client sensitivity to loss, termination phase evaluation, and overall treatment outcome.	Therapists	233	WAI-SF; RRI; COM	Therapists perceiving a stronger WA and RR during the termination phase were also likely to view the termination phase as effective and overall treatment as successful. Only therapists’ perceptions of the WA during the termination phase contributed to overall treatment outcome, when all three components were examined together.
Bhatia and Gelso (2018); USA	Cross-sectional	Examine the components of the tripartite model in terms of how they relate to one another and to the outcome of a psychotherapy session, from the therapists’ perspective.	Therapists	249	WAI-SR; RRI; SES	WA, RR, transference and countertransference contributed 27% of the variance in session outcome as rated by therapists of varying theoretical orientations. From the therapist’s perspective, the WA, RR and transference configuration were associated with session outcome. Only the RR and the WA predicted session outcome when all the components were looked at simultaneously in a regression model.
Pérez-Rojas and Gelso (2020); USA	Cross-sectional	Assess how acculturation may impact international student therapists’ experiences in cross-cultural counseling.	Therapists	104	WAI-S; RRI: SES	The self-reported quality of counseling relationships (RR and WA) and of counseling sessions with U.S. clients was unrelated to international counseling students’ levels of acculturative stress and perceptions of cultural distance.
Aafjes-van Doorn, Békés, and Prout (2020); USA	Cross-sectional	Explore therapists’ experiences of video therapy after switching from in-person to video sessions during the pandemic.	Therapists	141	WAI-SF; RRI	Higher rated WA and RR were associated with more positive attitudes towards video therapy. Scores on the RR and WA were positively correlated. Neutral WA online, albeit lower than those reported in previous studies on video therapy. Relatively strong RR online, similar to levels reported in studies of in-person therapy.

RR: Real Relationship; WA: Working Alliance; WAI-SF: Working Alliance Inventory – Short Form; WAI-SR: Working Alliance Inventory – Short Revised; SEQ: Session Evaluation Questionnaire; COM: Counseling Outcome Measure; SCL-90-R: Symptom Checklist-90-Revised; SOS-10: Schwartz Outcome Scale-10; OQ-45: Outcome Questionnaire-45; IIP: Inventory of Interpersonal Problems; SES: Session Evaluation Scale.

After close examination of all 23 studies, and because we wanted our analysis to be as thorough as the one performed by Gelso, Kivlighan, and Markin (2019), we decided that included studies should allow the calculation of the correlation between the real relationship and working alliance. Thus, studies that did not have that information (Pearson’s coefficient r) were excluded.

In addition, three studies examining the real relationship and working alliance were not included because their data sets partially overlapped with other studies that met inclusion criteria. In those cases, we used the studies that presented information such as the correlations between real relationship and working alliance instead of only including the means and standard deviations. For instance, a study by Gullo, Lo Coco, and Gelso (2012) was not included, because their sample is shared with the one of Lo Coco, Gullo, Prestano, and Gelso (2011), which contained the most inclusive data set. Similarly, the investigations conducted by Hill et al. (2014) and Pinto-Coelho, Hill, and Kivlighan (2016) were not included, because their sample is shared with the one of Shafran, Kivlighan, Gelso, Bhatia, and Hill (2017), which contained the most inclusive data set. Also, studies by Fuertes, Gelso, Owen, and Cheng (2013) and Morales, Keum, Kivlighan, Hill, and Gelso (2018) were excluded because they did not present correlations between real relationship and working alliance. The Baumann and Hill (2016) study was also excluded, because it only presented a correlation between real relationship and the Bond subscale of the Working Alliance Inventory (WAI). Finally, the Alessi, Dillon, and Van Der Horn (2019) study only presented correlations between the subscales of each measure (RRI and WAI) and not their totals, hence it was also excluded.

From the sixteen selected studies, along with data necessary for computing standardized effect sizes (Pearson’s r), we extracted the sample size, and information on the rater (client or therapist). Pearson’s correlation coefficient I was the effect size measure used in this research. We used the Comprehensive Meta-Analysis V3 (CMA) (www.meta-analysis.com) statistical software to conduct the analyses.

We followed the procedure of Gelso, Kivlighan, and Markin (2018), and computed Pearson’s r and 95% confidence intervals (CI) as summary statistics. For studies in which client and therapist ratings provided two correlations between working alliance and real relationship, these values were combined in the CMA software. A Fisher’s Z transformation of each correlation within the same study was calculated and then, the Fisher’s Z mean was back transformed to correlation. The heterogeneity among effect sizes in an analysis was assessed using the Q-statistic (assessing whether between-study heterogeneity exceeds that expected by chance alone). All analyses used random effects models.

Statistical analysis

Sixteen studies were included in the meta-analysis, which means an increase of 80% of studies compared to the meta-analysis of real relationship and working alliance performed by Gelso, Kivlighan, and Markin (2019) and an increase of 1119 participants.

This meta-analysis follows the criteria adopted by Gelso, Kivlighan, and Markin (2019). The omnibus effect size was significant (r = .66, 95% CI [.58, .73], p < .001, N = 2189 participants). Figure 2 displays the forest plot for this analysis. These results provide evidence for the conceptualization of real relationship and working alliance as sister concepts (Gelso, 2014), with medium to large associations, but not identical constructs.

Figure 2. Forest plot of effect sizes and confidence intervals for the meta-analysis of real relationship and working alliance.

“Box” size is relative to sample size, with larger boxes indicating a larger sample. “Favors A” indicates a negative correlation, whereas “Favors B” indicates a positive correlation. The last line of the table is the estimated results (random effects) for the meta-analysis.

One concern of publication bias is that some non-significant studies are missing from the analysis, and, if included, they would nullify the observed effect (Cooper et al., 2009). The number of studies that would be required to nullify the effect represent the fail-safe N. Because we need a large number of studies to nullify the mean effect, then there is no need to concern about publication bias. The fail-safe N was 5514.

Our results show that there was significant heterogeneity across the studies (Q[15] = 132.08, p = .000), and the extent of heterogeneity was high (I² = 89%), representing a high variability among studies. The Egger test also showed no evidence of publication bias (t = 1.45, p = .08).

Study characteristics

The included articles were published between 2005 and 2020. Most analyzed studies (N = 10) were cross-sectional, while the rest (N = 6) were longitudinal. Among the longitudinal studies, working alliance and real relationship were measured after each session (N = 4); on the third and eight sessions (N = 1); and on the third session and at termination (N = 1).

Because there is only one instrument to measure the real relationship, all studies used the RRI: the complete 24-item form was used more often (N = 11) than the short 12-item form (N = 5). On the other hand, although different measures have been used to study the working alliance, all studies included in our review used only the Working Alliance Inventory (WAI), the most commonly used measure. The Working Alliance Inventory – Short (WAI-S; Tracey & Kokotovic, 1989) was used in nine studies; whereas the Working Alliance Inventory – Short Revised (WAI-SR; Hatcher & Gillaspy, 2006) was used in six studies. Both are short 12-item versions that come from the complete 36-item WAI (Horvath & Greenberg, 1986, 1989). Only one study in our review used the complete form.

Regarding the raters, we observed heterogeneity, with real relationship and therapeutic alliance being rated by both therapists and clients in six studies, only by therapists in five and only by clients in another five.

The clients’ eligibility criteria included being 18 years old or older and having had at least some psychotherapy sessions: one session (N = 1); three sessions (N = 1); five sessions (N = 2); eight sessions (N = 2). Four studies recruited their participants before they started their psychotherapy sessions. One study was not clear on how many sessions the clients had completed. Although some studies were not specific on their clients’ eligibility criteria – aside from the minimum number of sessions – other criteria could include the nonexistence of psychotic or suicidal symptoms or substance abuse behaviors.

Overview of quantitative studies on real relationship and working alliance predicting outcome

Of the sixteen studies used on the meta-analysis, there were only ten studies which aimed to examine how working alliance and real relationship are related to outcome. We carried out a descriptive analysis of those ten studies to compare the real relationship and working alliance contribution to treatment outcomes.

First, we need to consider that outcome is not measured in the same way in all studies. Outcomes can be evaluated through session quality, client progress or symptomatic change. Seven outcome measures were used by the authors of the different studies: session quality (Session Evaluation Questionnaire, SEQ; Session Evaluation Scale, SES); client progress (Counseling Outcome Measure, COM; Schwartz Outcome Scale-10, SOS-10); and symptomatic evolution (Symptom Checklist-90-Revised, SCL-90-R; Outcome Questionnaire-45, OQ-45; Inventory of Interpersonal Problems, IIP). These measures can be completed by clients and/or therapists, and they can assess outcome after each session, at the beginning and end phases of psychotherapy (pretest-post-test change), or at a certain treatment phase. Despite all these differences between measures, they all address outcome, and we make no distinction between them.

Four studies (Fuertes et al., 2007; Gelso et al., 2005; Lo Coco, Gullo, Prestano, & Gelso, 2011; Marmarosh et al., 2009) concluded that the real relationship was a better predictor of outcome than working alliance. The Gelso et al. (2005) study revealed a small and non-significant association between working alliance and the Depth and Smoothness subscales of the SEQ (Depth: r = .14; Smoothness: r = .16). On the other hand, the RRI had a positively significant association with the subscales of the SEQ (Depth: r = .36, p < .01; Smoothness: r = .43, p < .01). Fuertes et al. (2007) found that client ratings of the real relationship predicted 14% of additional variance in their ratings of psychotherapy progress above and beyond client ratings of attachment, working alliance, and therapist empathy (adjusted R² = .54). Therapist ratings of the real relationship explained an additional 5% of variance in therapist ratings of client progress above and beyond attachment and their ratings of the working alliance. However, this result was only marginally significant (β = .29, p < .058). Client and therapist ratings of the real relationship, and not their ratings of the working alliance, predicted their ratings of client progress. In the study by Marmarosh et al. (2009), hierarchical multilevel regression revealed that therapist-rated real relationship was the only significant predictor of post-treatment symptoms. In addition, therapists’ perceptions of realism but not genuineness, accounted for a significant amount of variance in client-rated therapy outcomes. Client-rated real relationship, especially the genuineness subscale, predicted outcome in the study conducted by Lo Coco, Gullo, Prestano, and Gelso (2011) and added to the working alliance effect in predicting outcome a significant increase in the explained variance (adjusted R² increased from .10 to .38). The working alliance did not seem to relate to treatment outcome, except for the client-rated bond element of the working alliance.

Only one study (Bhatia & Gelso, 2017) revealed a greater role of working alliance in predicting outcome. Real relationship, working alliance, and transference were examined through the therapists’ perspectives during the termination phase of therapy, and were related to overall treatment outcome and other variables. Results showed that the three relational components together predicted 19% of the variance in overall treatment outcome (R² = .19, F_{(3, 216)} = 18.67, p < .01) during the termination phase. However, only therapist-rated working alliance significantly predicted treatment outcome (B = .35, p < .01).

Four studies found no differences between real relationship and working alliance in their prediction of outcome (Bhatia & Gelso, 2018; Kivlighan, Kline, Gelso, & Hill, 2017; Owen, Tao, Leach, & Rodolfa, 2011; Pérez-Rojas & Gelso, 2020). Owen, Tao, Leach, and Rodolfa (2011) found a positive significant association between client-rated psychological well-being and real relationship (r = .27, p < .001) and working alliance (r = .34, p < .001); Pérez-Rojas and Gelso (2020) also showed a significant positive association between therapist-rated session quality and real relationship (r = .47, p < .001) and working alliance (r = .42, p < .001). Kivlighan, Kline, Gelso, and Hill (2017) examined how congruence and discrepancy in clients’ and therapists’ ratings of the real relationship and working alliance were related to client-rated session quality. Their main finding revealed that, for both clients and therapists, session quality was highest when the combined real relationship and working alliance ratings were high and lowest when combined ratings were low. Bhatia and Gelso (2018) conducted a simultaneous regression analysis to examine how working alliance, real relationship, negative transference, and countertransference behaviors, as perceived by therapists, contributed to therapist ratings of session outcome. Results indicated that the four components together predicted 27% of the variance in session outcome (Adjusted R² = .27, F₍₂₃₇₎ = 23.10, p < .01). Further examination of the regression model revealed that only the real relationship and working alliance significantly predicted session outcome after adjusting for all components of the tripartite model (RRI: B = .73, p < .01; WAI: B = .98, p < .01).

Lastly, only one study showed differences between real relationship and working alliance when the raters were different. Kivlighan, Hill, Gelso, and Baumann (2016) examined the associations between the dyad on their ratings of real relationship, working alliance, session quality and outcome using an actor – partner interdependence model (APIM) procedure, and demonstrated that clients’ and therapists’ evaluations of the working alliance and the real relationship were significantly related to their own evaluations of session quality. For therapist session outcome, the therapist working alliance effect was twice as large as the therapist real relationship effect. By contrast, for client session outcome, the client real relationship effect was twice as large as the client working alliance effect. These results suggest that while therapists give more weight to the working alliance, clients pay more attention to the real relationship when evaluating sessions.

Moreover, to supplement the abovementioned overview of quantitative studies, we performed a comparison of correlations from dependent samples using the studies that assess the relationship between real relationship and outcomes and working alliance and outcomes. We also computed the correlations’ weighted mean using the Fisher’s Z transformation (Psychometrica, n.d.) that allows the consideration of each sample size.

Results present in Table 2 showed that one study significantly favors the real relationship-outcomes relationship (Gelso et al., 2005) and two of them the working alliance-outcomes relationship (Bhatia & Gelso, 2017, 2018). However, considering the total weighted mean, results do not favor either the relationship between real relationship and outcomes or between working alliance and outcomes (Z = .623; p = .267; Table 2).

Table 2. Comparison of correlations between working alliance and outcomes and real relationship and outcomes from dependent samples.

Study (Author, year)	Raters	N	r WA-outcomes^1	r RR-outcomes^2	Z	p
Gelso et al. (2005)	Therapists (SEQ – Depth)	92	0.14	0.36	−2.112	0.017
	Therapists (SEQ – Smoothness)		0.16	0.43	−2.647	0.004
Fuertes et al. (2007)	Therapists	59	0.27	0.36	−0.72	0.236
	Clients	59	0.14	0.29	−1.518	0.064
Marmarosh et al. (2009)	Therapists	21	0.02	0.01	0.051	0.479
	Clients	31	0.07	0.2	−1.361	0.087
Lo Coco, Gullo, Prestano, and Gelso (2011)	Therapists	4	0.14	0.14	0.409	0.341
	Clients	50	0.18	0.21	−0.215	0.415
Owen, Tao, Leach, and Rodolfa (2011)	Clients	176	0.34	0.27	1.302	0.096
Kivlighan, Hill, Gelso, and Baumann (2016)	Therapists	23	0.61	0.51	0.693	0.244
	Clients	74	0.61	0.5	1.473	0.07
Bhatia and Gelso (2017)	Therapists	233	0.45	0.29	2.785	0.003
Bhatia and Gelso (2018)	Therapists	249	0.5	0.42	1.687	0.046
Pérez-Rojas and Gelso (2020)	Therapists	104	0.42	0.47	−0.677	0.249
Total Weighted Mean		1175	0.37	0.356	0.623	0.267

¹Correlation values between Working Alliance and outcomes; ²Correlation values between Real Relationship and outcomes.

Discussion

The first goal of this study was to perform a meta-analysis to verify the association between real relationship and working alliance, adding seven more studies than the previous meta-analyses (Gelso, Kivlighan, & Markin, 2019). Secondarily, in both descriptive and quantitative analyses, we sought to compare how the working alliance and real relationship are associated to outcomes and under which conditions.

Our results revealed a significant omnibus effect on the association between real relationship and working alliance that was slightly greater than the one found by Gelso, Kivlighan, and Markin (2019), demonstrating a slightly larger overall correlation between these concepts. These results confirm the existence of a large association between real relationship and working alliance.

But to what do we attribute this association? Kelley, Gelso, Fuertes, Marmarosh, and Lanier (2010) investigated if the real relationship and working alliance were that different. To do so, they examined if the RRI-C correlated differently with the three WAI subscales. They discovered that the bond subscale of the WAI was more highly correlated with the real relationship (r = .81) than were the other components of the WAI (rs of .66 and .65 for goals and tasks, respectively). This makes sense given that the bond subscale may be easily confused with the personal bond that the real relationship represents, especially for clients. In fact, Gelso (2014) had already noted that three of the four items of the WAI bond subscale of the 12-item measure (Hatcher & Gillaspy, 2006) tap personal feelings between therapist and client (“I believe my therapist likes me”/“I believe my client likes me”; “I feel that my therapist appreciates me”/“I appreciate my client as a person”; “My therapist and I trust one another”/“My client and I have built a mutual trust”) and only one taps the work collaboration (“I’m confident in my therapist’s ability to help me”/“I’m confident in my ability to help my client”). Although not possible in our theorical review, we consider it important to investigate further the contribution of the WAI bond subscale in the association between real relationship and working alliance, since it may be a factor that reinforces or even underlies the conceptual confusion between the concepts.

After confirming the association between the constructs, results from studies allowed us to observe how they are related to outcomes. When comparing the correlations between working alliance and outcomes and real relationship and outcomes from the whole samples of these studies, we found that the two constructs related equally to outcome. This seems to underscore that both constructs are important and seem to benefit from each other’s effect, suggesting that the real relationship is as good a predictor of the outcome as the working alliance.

Although we did not explore a possible moderating effect of the rater, Kivlighan, Hill, Gelso, and Baumann (2016) revealed that each member of the therapeutic dyad may attribute different importance to the components of the therapeutic relationship. Future research could further explore the underlying reasons for the difference between clients’ and their therapists’ emphasis on the importance of the real relationship and the working alliance. Qualitative research might be particularly useful in this effort.

Regarding the studies included in the meta-analysis, only one was of online therapy (Aafjes-van Doorn, Békés, & Prout, 2020) as it reports the transition to online therapy during the COVID-19 pandemic; however, the working alliance and the real relationship were established in person, making it equivalent to the other studies. This study reported that therapists who transitioned to online therapy perceived a sufficient working alliance and a strong real relationship. In fact, a systematic review performed by Norwood, Moghaddam, Malins, and Sabin‐farrell (2018) reinforced that there is a good working alliance and outcome in videoconferencing psychotherapy. However, online therapy may have a slight negative effect on the therapeutic relationship since patient and therapist are not as synchronized during psychotherapy as they would be in face-to-face therapy. The nonverbal processes (vocal pitch, head movements, and whole-body movements) that characterize the synchrony are elementary for the alliance establishment (Koole & Tschacher, 2016). These processes are not as clear in online therapy, which can negatively affect both working alliance and real relationship. For this reason, future studies should address the distinct effects of face-to-face and online therapy on the establishment and maintenance of both the real relationship and the working alliance.

In sum, both the conceptualization of the constructs and the empirical results of this study support Gelso’s characterization of real relationship and working alliance as “sister concepts”. We believe, however, that they might be at a critical development phase of their “adolescence”, in that research is beginning to show different patterns in the relationship of the two constructs to treatment effectiveness. It appears that real relationship may prove to be as good predictor of outcome as the working alliance, but there are far too few studies to allow for firm conclusions. Secondly, it seems like the association between real relationship and working alliance with outcome may be influenced by the rater: therapists consider working alliance a better predictor, while clients think that real relationship is the one with a better predictive value. Again, it means that much more empirical work is needed before definitive conclusions may be drawn.

We also need to add that both the real relationship and the working alliance are known as common factors. That is, both may be seen as significant contributors to psychotherapy success across differing theoretical approaches. Although studies until now have focused on analyzing the isolated contribution of each variable, recent scientific appeals highlight the need to assess how the interaction between the distinct common factors contributes to outcomes (Cuijpers, Reijnders, & Huibers, 2019; Norcross & Lambert, 2018), and how this interaction changes over time in the therapeutic processes (Altman, Shapiro, & Fisher, 2020; Ramseyer, Kupper, Caspar, Znoj, & Tschacher, 2014).

Until now, more attention has been paid to the working alliance. We could say that this is understandable, given the fact that it is easier to work on concrete dimensions, such as the goals of therapy and the tasks required to achieve those goals. The real relationship involves personal elements in the therapeutic relationship, while the working alliance concerns an explicit idea of negotiation between therapist and client. Still, our evidence recommends that therapists should be able to develop a real relationship with their clients and be more aware of the quality of this relationship.

The present study has extended the findings of previous meta-analyses. However, there are substantial limitations to note. One significant limitation concerns the small number of studies regarding real relationship, which may have influenced the effect sizes of our meta-analysis. Only with the validation of the first measures of real relationship has it been possible to conduct empirical studies on this subject, so there are still relatively few investigations that explore the construct. A second limitation is related to the absence of a moderator analysis to understand possible causes of heterogeneity among study conclusions. In addition, another limitation concerns with the risk of confirmation bias related to the interpretation of results favoring the real relationship, since nearly all the studies to date that explore this concept have been conducted by Gelso and his colleagues and students. Gelso (2019) himself has already mentioned that since the studies have been conducted by the same researchers, there may be a bias that favors the results related to the real relationship dimension. Therefore, the findings related to the real relationship should be interpreted with caution, and the concept should be further investigated by other research teams.

More broadly, we recommend that future studies on the current topic continue the exploration of the overlap between the two sister concepts. Additional investigations should include repeated measures designs which allow for the use of robust statistical procedures considering variables’ temporal variations. Recently, Flückiger et al. (2022) applied a longitudinal design to explore changes in clients’ alliance over the course of treatment and observed that, in good outcome cases, alliance concept evolved from the three dimensions (tasks, goals and bond) to a single factor. Applying this study approach to discriminate the two sister concepts, it would be important to closely examine the links between the subscales of working alliance and the subscales of real relationship over time.

In our analysis, the real relationship was examined as a whole, combining the theorized elements of realism and genuineness. Although there may be doubts about this integration, it seems to make sense, both theoretically and from a measurement standpoint. From the inception of the real relationship construct (Greenson, 1967; Gelso, 2011), realism and genuineness have been viewed as closely intertwined and measures of the real relationship witness a very strong correlation between realism and genuineness for both client (r = .75, Kelley, Gelso, Fuertes, Marmarosh, & Lanier, 2010) and therapist ratings (r = .80, Gelso et al., 2005). Still, it is likely that there are variables with which realism and genuineness interrelate differentially (e.g. Marmarosh et al., 2009). Thus, it would be fruitful to examine how realism and genuineness may differentially relate to other variables.

Other questions that could be posed include how the transference-countertransference configuration interacts with the real relationship and the working alliance, and to what extent these three components have a differentiated contribution to the outcomes when examined together. It would be important, as hypothesized by Gelso (2013), to assess how certain transference-countertransference patterns relate to the alliance and the real relationship to help shed light on the interrelationships between the three components of the therapeutic relationship.

Despite the need of further research, the present study contributes to the current debate concerning the association between real relationship and working alliance through a psychometric analysis of comparison between correlations which proves that real relationship and working alliance have equal predictive value to explain therapeutic outcomes. This finding reinforces the importance of further exploring the specific contribution of each variable, in addition to the overlapping element, since the specific contribution may be associated with different factors, such as the evaluator.

Practical implications

This study sought to explore the meta-analytic association between real relationship and working alliance. The results confirm a large association between real relationship and working alliance, showing that real relationship is as good a predictor of outcomes as working alliance. Our findings highlight the need of considering both the shared and the specific contribution of these common factors and the need of further investigating interrelationships between them using longitudinal and time-series designs.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

The work was supported by the Fundação para a Ciência e a Tecnologia [CIP - Refª UIDB/PSI/04345/2020].

Notes on contributors

Ana Marta Vaz

Ana Marta Vaz has an MSc in Clinical and Health Psychology at the Universidade do Algarve and is an effective member of the Order of Portuguese Psychologists (OPP).

Laura Inês Ferreira

Laura Inês Ferreira is currently a second-year doctoral scholarship student on the PhD in Psychology at the Universidade do Algarve and integrates the UAlg Research Unit of Psychology Research Center of Universidade Autónoma de Lisboa (CIP-UAL). She obtained an MSc in Clinical and Health Psychology in 2018 and is an effective member of the Order of Portuguese Psychologists (OPP). Her research interests include psychodynamic psychotherapy process and outcome, therapeutic alliance, emotional experience, and the change mechanisms.

Charles Gelso

Charles Gelso is a Professor Emeritus and Senior Lecturer at the University of Maryland and an academic researcher. He received his doctoral degree in Counseling Psychology from Ohio State University in 1970 and started working as a full-time professor in 1982. He have co-authored more than 165 scientific publication(s), published one book and chaired several dissertation committees. His research and clinical interests focus on the patient-therapist relationship in psychotherapy. Charles Gelso have also conducted individual psychotherapy throughout all his career, mostly based on the psychodynamic approach.

Luís Janeiro

Luís Janeiro is an Assistant Professor at the Department of Psychology and Educational Sciences of Universidade do Algarve and a Researcher at UAlg Research Unit of Psychology Research Center of Universidade Autónoma de Lisboa (CIP-UAL). He has PhD in Clinical and Health Psychology and is an effective member of the Order of Portuguese Psychologists (OPP). His main research interests are focused on change processes in psychotherapy and treatments for substance use disorders. Luís Janeiro is also the director of the UAlg Psychology Service and coordinator of the Clinical and Health Psychology, where he develops his clinical activity as a psychotherapist and supervises professional internships, based on his training in Intensive Short-Term Dynamic Psychotherapy.