The working alliance in individual therapy

Full text

Alliance in Individual Psychotherapy
Adam O. Horvath
Simon Fraser University
A. C. Del Re
University of Wisconsin–Madison
Christoph Flu¨ckiger
Universita¨t Bern and University of Wisconsin-Madison
Dianne Symonds
Kwantlen Polytechnic University
This article reports on a research synthesis of the relation between alliance and the outcomes of individual
psychotherapy. Included were over 200 research reports based on 190 independent data sources, covering
more than 14,000 treatments. Research involving 5 or more adult participants receiving genuine (as
opposed to analogue) treatments, where the author(s) referred to one of the independent variables as
“alliance,” “therapeutic alliance,” “helping alliance,” or “working alliance” were the inclusion criteria.
All analyses were done using the assumptions of a random model. The overall aggregate relation between
the alliance and treatment outcome (adjusted for sample size and non independence of outcome
measures) was r⫽.275 (k⫽190); the 95% confidence interval for this value was .25–.30. The statistical
probability associated with the aggregated relation between alliance and outcome is p⬍.0001. The data
collected for this meta-analysis were quite variable (heterogeneous). Potential variables such as assess-
ment perspectives (client, therapist, observer), publication source, types of assessment methods and time
of assessment were explored.
Keywords: therapeutic alliance, psychotherapy relationship, working alliance, meta-analysis, psychotherapy
outcome
The concept of the alliance is currently one of the most intensely
researched subjects in the psychotherapy research literature. A
search of the electronic databases in 2009 has yielded over 7000
items using the key words: alliance, helping alliance, working
alliance, and/or therapeutic alliance. The popularity of the alliance
concept in the research community can be traced back, in part, to
the interest generated by the meta-analyses published in the 1970s
(e.g., Luborsky, Singer, & Luborsky, 1975; Smith & Glass, 1977;
Stiles, Shapiro, & Elliot, 1986). These studies reached the general
conclusion that diverse therapies provided similar beneficial ef-
fects to psychotherapy clients. This finding has strongly encour-
aged the research community to search for the common factors
shared by different psychotherapies. The relationship between
therapist and client was an obvious candidate for such generic
factor.
Another important source of the growing interest in the therapy
relationship was the work of Rogers and his colleagues (Rogers &
Wood, 1974). By applying rigorous empirical methods to the
examination of the role of the therapist-offered facilitative condi-
tions in person-centered treatment, they moved the concept of the
therapeutic relationship to the center of the research agenda.
A third important precursor can be traced back to the 1930s: A
growing curiosity and interest in the integration of diverse theories
of psychotherapies (Frank & Frank, 1991; Rosenzweig, 1936). The
desire to reconcile some conflicting therapeutic methods and their
underlying theories eventually led to the founding of the Society
for the Exploration of Psychotherapy Integration (SEPI) in 1983.
Consistent with this trend, many psychotherapists in North Amer-
ica started to reject the strict boundaries of classical theories and
became increasingly interested in utilizing a variety of effective
methods, irrespective of the theories from which these strategies
originated from. The field was moving from theoretical monism to
an eclectic pragmatism. The value of aspects of therapist-client
relationship (e.g., alliance) found ready acceptance among those
committed to psychotherapy integration (Goldfried, 1980). But
perhaps the most potent force responsible for the sustained growth
of interest in the alliance was the consistent finding of a moderate
but robust relationship between the alliance and treatment outcome
across a broad spectrum of treatments in a variety of client/
Adam O. Horvath, Faculty of Education and Department of Psychology,
Simon Fraser University, Burnaby, British Columbia, Canada; A. C. Del
Re, Department of Counseling Psychology, University of Wisconsin–
Madison; Christoph Flu¨ckiger, Department of Clinical Psychology and
Psychotherapy, University of Bern, Switzerland, and Department of Coun-
seling Psychology, University of Wisconsin–Madison; and Dianne Sy-
monds, Department of Nursing Science, Kwantlen Polytechnic University,
Vancouver, British Columbia, Canada.
This article is adapted, by special permission of Oxford University
Press, from a chapter of the same title by the same authors in J. C. Norcross
(Ed.), 2011, Psychotherapy relationships that work (2nd ed.). New York:
Oxford University Press. The book project was cosponsored by the APA
Division of Psychotherapy.
We acknowledge the contribution of F. Groenewold, PhD and our
research assistants V. Velasco, J. Wells, S. Nand, and R. Yacoub. We also
wish to express our thanks for the statistical advice of B. Wampold, Ph.D.
Correspondence concerning this article should be addressed to Adam O.
Horvath, Simon Fraser University, 8888 University Way, Burnaby, BC
V5A 1S6, Canada. E-mail: horvath@sfu.ca
Psychotherapy © 2011 American Psychological Association
2011, Vol. 48, No. 1, 9–16 0033-3204/11/$12.00 DOI: 10.1037/a0022186
9

problem contexts (Horvath & Symonds, 1991; Martin, Garske, &
Davis, 2000; Horvath & Bedi, 2002).
In this paper, we present a new research synthesis of the relation
between the alliance and psychotherapy outcome in individual
therapy. Using the accumulated research data we also explore the
role of several potential moderators that could impact this rela-
tionship.
Definitions and Measures
History and Definitions
The concept of the alliance (but not the term itself) dates back
to the middle period of Freud’s writings during which he recon-
sidered and elaborated the role and function of transference in
psychotherapy. In some of his writings (Freud, 1912/1958; 1913),
he noted the apparent paradoxical situation patients find them-
selves in the beginning of treatment; the therapy process itself
activates the client’s defenses which should make the patient flee
the therapeutic situation, yet, in successful treatments, clients
persist to collaborate with the therapist in unearthing disturbing
material. As a solution to this contradiction, he proposed the
presence of a positive or “unobjectionable” transference which
binds the client to the person of the therapist and assists the patient
to remain in treatment despite the increased level of anxieties. This
concept was subsequently further elaborated by Sterba (1934),
Zetzel (1956), and Greenson, 1965). The term alliance was coined
by Zetzel (1956) and conscious aspects of the concept of the
alliance was emphasized and elaborated by Greenson (1967).
During the 1970s efforts were made to extrapolate and extend
the concept of the alliance from its psychodynamic roots to en-
compass the collaborative relational elements in all helping en-
deavors: Luborsky (1976) proposed an extension of Zetzel’s
(1956) and Stone’s (1961) concept of the alliance. He suggested
that the alliance between therapist and client developed in two
phases: The first phase, involved the client’s belief in the therapist
as a potent source of help, and the therapist providing a warm,
supporting, and caring relationship. This level of alliance results in
a secure holding relationship within which the work of the therapy
can begin. The second phase, Type II alliance, involved the client’s
investment and faith in the therapeutic process itself, a commit-
ment to the core concepts undergirding the therapy (e.g., nature of
the problem, value of the exploratory process) as well as a willing
investment of her or himself to share the ownership for the therapy
process.
Bordin (1975, 1989, 1994) proposed a somewhat different pan-
theoretical alliance concept. He named it the working alliance. His
concept of the alliance were based on Greenson’s (1965) ideas, but
departed from the psychodynamic premises even more clearly than
Luborsky’s did. For Bordin, the alliance was centrally the achieve-
ment of a collaborative stance in therapy and its development was
fostered by three processes: agreements on the therapeutic goals;
consensus on the tasks than make up therapy; and a bond between
the client and the therapist. He predicted that different therapies
would emphasize different aspects of the alliance. Bordin (1994)
also proposed that, as therapy progresses, the strength of the
working alliance would build and ebb in the normal course of
events, and that the repair of these stresses in the alliance offers
potent therapeutic possibilities and make a direct contribution to
clients’ change.
The most distinguishing feature of the modern pan-theoretical
reconceptualization of the alliance is its emphasis on collaboration
and consensus (Bordin, 1980; Hatcher, Barends, Hansel, & Gut-
freund, 1995; Luborsky, 1976). In contrast to previous formula-
tions that emphasized either the therapist’s contributions to the
relationship (i.e., Rogers & Wood, 1974) or the unconscious
distortions of the relation between therapist and client (i.e., Freud,
1912), the “new” alliance concept emphasized the conscious as-
pects of the relationship (as opposed to unconscious processes) and
the achievement of collaborative, “work together” aspects of the of
the relationship.
However, neither of these early advocates of the pan-theoretical
alliance construct chose to offer a concise definition of the term.
This lack of a precise consensual definition has, on one hand, made
it easier for researchers and clinicians of diverse theoretical frame-
works to embrace the term and integrate it within their specific
conceptualization of the therapy process. But on the other hand,
this “creative ambiguity” led to some problematic developments in
the research literature. The most immediate and direct conse-
quence was that a number of alliance measures that were devel-
oped in parallel between 1978 and 1986 did not share a clear
common point of reference. In the absence of a shared definition
these —and subsequent— alliance measures de facto define what
the researcher means by the term “alliance.” While there are some
important shared aspects across many of the alliance measures
(e.g., Bordin, 1980, 1989; Gaston et al., 1995; Hatcher & Barends,
2006; Horvath & Luborsky, 1993), there are also nontrivial dif-
ferences among authors about the meaning of the term alliance
(e.g., Psychotherapy: Therapy, Research, Practice, Training,
2006, 43[3]). In terms of the research synthesis we present in this
report, it is important to emphasize that what we know about the
alliance and its relation to outcome and other therapy variables has
been gleaned from studies which, in practice, define the alliance by
the diverse instruments used to measure it.
Alliance Measures
In the remainder of this article we refer to the alliance in the
singular. However within the 201 studies in our collection of data,
over 30 different alliance measures were used—not counting dif-
ferent versions of the same instruments. Similar to previous re-
ports, the four “core measures”: California Psychotherapy Alliance
Scale, (CALPAS), Helping Alliance Questionnaires (HAq), Van-
derbilt Psychotherapy Process Scale (VPPS), and Working Alli-
ance Inventory (WAI) accounted for approximately 2/3 of the data.
In research on the shared factor structure of the WAI, CALPAS
and HAq, the concept of “confident collaborative relationship”
was identified as the central common theme (Hatcher & Barends,
1996; Hatcher et al., 1995). Each of these four instruments has
been in use for over 20 years and has demonstrated an acceptable
level of internal consistency (Martin et al., 2000). However, the
shared variance, even among these “core” measures, has been
shown to be less than 50% (Horvath, 2009).
Fifty-four of the research reports in our data set used less well
validated instruments or assessment procedures; the relation of
most of these measures to the core instruments, or to each other,
are not well documented, and sometimes nonexistent. Later in this
10 HORVATH, DEL RE, FLU
¨CKIGER, AND SYMONDS

paper, when we report on the analysis of potential moderators, the
less often used measures (nofuseⱕ3) are merged into one
category: “Other.” In this “Other” category are some newer alli-
ance measures with relatively few administrations; measures de-
veloped for the specific investigation; and instruments originally
developed for relationship constructs other than the alliance. Add-
ing to the diversity of assessment perspectives is the fact that the
four core instruments currently exist in a number of different forms
(e.g., short versions, observer versions, versions specific to context
and/or application, translations). The relation of these modified
instruments to the original versions is not always well documented.
As we noted, the diversity in the “de facto” definition of the
alliance that has emerged via the use of a variety of assessment
measures has became an important source of variability across
studies. The consequences of this diversity will be discussed in
more detail in the moderator section of this report.
Clinical Examples
The alliance represents an emergent quality of partnership
and mutual collaboration between therapist and client. As such,
it is not the outcome of a particular or typical intervention. Its
development can take different forms and maybe achieved
quickly or nurtured over a longer period of time depending on
the kind of therapy and the stage of treatment (Bordin, 1994).
The following is an excerpt intended to provide a flavor of
therapist-client interactions that likely contribute to the devel-
opment of the alliance.
Client (C), therapist (T), taken from the 5th session, 20 min into
the interview
C: Well aren’t you going to ask me what this reminds me of?
T: You think I should?
C: You do; always.
T: Because we agreed that looking at connection between past
relationship patterns and how you and [name] are getting on
is...
C: {voice over} Yes, unfinished business . . . and all that.
T: It may be that there is a pattern here, which would be
useful to explore and understand better. Once we understand
it, we can recognize it, and perhaps prevent a replay of the
same old grooves . . . [pause10 sec]
T: I said that “we agreed” that this is the way to go, but I get
the sense that you may not be convinced that’s so . . . it is
such a good idea.
C: Look, I mean . . . you are the therapist and I keep fucking
up with my “old lady”. So I guess I better start thinking &
talking about these patterns . . . I wish there was a pill or
electric shock therapy to . . ., it would be faster. [sigh]
T: Maybe we better take a step back. I am a therapist, but I
can’t give you a pill or shock you to fix you. And looking for
these unfinished patterns don’t seem to make an awful lot of
sense to you . . . right now. But I hear you are willing to be a
“good client”.
C: But this not what it is about, about me being good, I mean,
right?
T: What would you say if you were not a “good patient”?
Would you rebel?
C: I guess I might . . . It’s crazy you know, before I got
married I was a pretty wild dog . . . long hair, motorcycles,
some pretty crazy stuff.
T: So, What happened? Where did the “crazy you” go? What
did you do with him?
C: Married, good job, slick house, nice kids, you know . . .
T: You think I might meet this character? He seems to have
been shut up but not forgotten . . . He might have something
interesting to say . . .
C: I might be a little afraid of my old self . . . But [with
different voice]: Doc, I’m trash, my old man was trash, but he
put his money in good booze; not in psychiatrists’ pockets!
T: He did not have much faith in this therapy business
C: Yeah, of course you should not let him write the check for
the session; it would for sure bounce . . . [both laugh]
In the above excerpt the therapist starts off defending his idea of
what is useful to do in therapy, but when he becomes aware of the
client’s ambivalent feelings about dealing with the past—and
possibly about being in therapy— he drops his previous agenda
and demonstrates his commitment to find a way of working
collaboratively with his patient. We did not include the material on
how the issue of dealing with past relationship patterns was re-
solved. What is important from the perspective of the building of
the alliance is that the therapist stepped back from pursuing his
original agenda and prioritized the negotiation of a collaborative
therapy relationship. We believe that the therapist response builds
the alliance on two levels. In the near term, it leads to the selection
of a method of intervention that is congruent with the client’s
resources and expectations. On the longer term, the therapist
makes the “metapoint” that successful treatment needs the client’s
full participation and collaboration and that he, the therapist, is
prepared to be responsive and adopt his method of treatment in
order to achieve a high level of mutuality. This brief excerpt also
illustrates that the concept of the alliance unites the notions of
interventions/strategies and the development of the relationship in
therapy. Alliance is built by doing the work of therapy collabora-
tively.
Meta-Analytic Review
Sources of Data
This research synthesis is a fourth meta-analysis published since
1991 summarizing the relation of the alliance to psychotherapy
11
SPECIAL ISSUE: ALLIANCE IN INDIVIDUAL PSYCHOTHERAPY

outcome. For identifying studies published between 1973 and
2000, we relied on data from previous analyses (Horvath & Sy-
monds, 1991; Martin et al., 2000; Horvath & Bedi, 2002) but the
effect sizes (ES) where recalculated (using more up-to-date meth-
ods) for all but 10 of the oldest unpublished studies which were no
longer available.
To locate data from the years 2000 to 2009, we first searched the
electronic databases (PsycINFO/EBSCO) using the same key-
words as the Horvath and Bedi (2002) analysis. Next we cross-
referenced the bibliography of studies included in the analysis. The
criteria for inclusion in this report were: (1) the study author
referred to the therapy process variable as “alliance” (including
variants of the term); (2) the research was based on clinical as
opposed to analogue data; (3) five or more adult patients partici-
pated in the study, and; (4) the data reported were such that we
could extract or estimate a value indicating the relation between
alliance and outcome.
In contrast to previous meta-analyses, the literature search was
extended to material available in Italian, German, or French, as
well as English. To accomplish this, a search was conducted of the
German language database (PSYNDEX) using the same inclusion
criteria as the English language searches. Seventeen German lan-
guage manuscripts contained usable alliance-outcome data and
were included in the analysis. For the French and Italian literature,
we searched in PsycINFO with the additional keywords French
OR Francais OR Italian OR Italiano This search yielded two
usable items. Twenty-six Italian manuscripts were located; of these
14 were published in English journals, none of the Italian-only
papers had usable data. In total, 19 research reports unavailable in
English were included in the analysis.
The data on which our analysis is based include both published
(158) and unpublished (53) research. The published research ap-
peared overwhelmingly (153) in peer-reviewed journals, 5 studies
came from book chapters, while 43 items came from unpublished
(mostly dissertations) sources. The later represent a significant
increase in the proportion of unpublished research in the current
data compared to previous meta-analyses. In total the data in this
meta-analysis captured information based on over 14,000 treat-
ments.
Of the 201 research reports in this meta-analysis, 39 manuscripts
were based on a shared data sets; that is, two or more reports
provided alliance-outcome information derived from common
pool of clients. As a result some of these reported effect sizes were
not independent. In addition, 10 research publications reported
multiple alliance outcome relations based on two or more inde-
pendent samples. As a first step, we computed the ESs associated
with each of the 201 manuscripts, but the aggregated effect sizes,
and all of the calculations presented below, were adjusted for
shared (non independent) data, and are based on the 190 indepen-
dent effects sizes.
The number of studies in the current study is roughly double the
size of the data available for the previous (Horvath & Bedi, 2002)
meta-analysis. The growth in the literature over the past decade
means that not only that there are more studies available for
analysis, but also that there is a significant increase in the types of
therapies, treatment contexts, client problems, and research de-
signs captured by this meta-analysis. Given that we have also
included studies only available in languages other than English
1
it
seems fair to claim that the data is a reasonable representation of
the research on the alliance outcome relation to date.
Methods of Analysis
All numerical estimates were calculated using restricted maxi-
mum likelihood (random-effects) model (Viechtbauer, 2005). The
reasons for this were twofold: First, using the alternative (fixed
effects) model we would have had to “. . . assume homogeneity of
underlying treatment effects across studies [and this would have]
lead to substantial understatement of uncertainty” (National Re-
search Council, 1991, p. 187). Second, the random-effects model,
apart from requiring fewer assumptions, yields a more conserva-
tive estimate, hence leads to safer, more trustworthy, conclusions
(Cooper, Hedges, & Valentine, 2009; Hunter & Schmidt, 2004).
Using a random effects model we assumed that the studies in our
data set are drawn from a population of studies and thus the results
of our analyses are generalizable to the larger universe of studies.
In many studies, there were a number of different outcome
measures. In order to account for the dependencies among out-
come measures, due to multiple within-study ESs, we used Hunter
and Schmidt’s (2004) aggregation procedures to obtain one corre-
lation effect size per study. These procedures take into account the
correlation among within-study outcome measures, and thus yield
a more precise estimate of the population parameter. In cases
where the studies did not provide actual correlations among out-
come measures, the estimate of between outcome measure corre-
lation was set to .50 (Wampold et al., 1997).
To correct for the non-normality of the distribution of the
correlation coefficients, for categorical and continuous moderator
analyses, all correlations were transformed to a Fisher’s z (Fisher,
1924) and then transformed back to rfor interpretive purposes. In
cases where the primary study reported more than one level of a
categorical variable (e.g., reporting both early, mid, and late alli-
ance and outcome correlations), dependencies at the moderator
level were accounted for by randomly selecting one within-study
level per study. All computations for this meta-analysis were
conducted using the MAc (Del Re, 2010) and RcmdrPlugin.MAc
(Del Re & Hoyt, 2010) meta-analysis packages for the R statistical
software program (R Development Core Team, 2009).
Results
The aggregate effect size (ES), for the 190 independent alliance/
outcome relations was r⫽.275. The 95% confidence interval of
this averaged ES ranged from .249 to .301. This aggregated value
is adjusted for sample size, as well as the intercorrelation among
outcome measures. The magnitude of the relationship we found in
the current meta-analysis is a little larger but similar to the values
reported in previous research (Horvath & Symonds, 1991 r⫽.26,
k⫽26; Martin et al., 2000, r⫽.22, k⫽79; Horvath & Bedi,
2002, r⫽.21, k⫽100). The median effect size of ESs of the
current data set was .28 (not adjusted for sample size) suggesting
that the group of effect sizes we collected was not strongly skewed.
The overall effect size of .275 is statistically significant at p⬍
1
Nothwithstanding this extended effort, the coverage is not fully com-
prehensive. We did not have the resources to search the Asian languages,
nor the European literature beyond English, German, Italian, and French.
12 HORVATH, DEL RE, FLU
¨CKIGER, AND SYMONDS

.0001 level indicating a moderate but highly reliable relation
between alliance and psychotherapy outcome.
Threats to Validity
The estimate of an aggregated effect size in a meta-analysis is
potentially vulnerable to systematic publication bias; the databases
we searched may be missing research reports that were not pub-
lished because these investigators failed to find a statistically
significant relation between the alliance and outcome. This is the
“file drawer problem” (Sutton, 2009). While it is not possible to
locate the unpublished studies that languish in the authors’ file
drawer, it is possible to compute a failsafe number, that is, an
estimate of the number of studies with an ES ⫽0.0 that would
have to be out there but missing from our data in order make the
aggregate effect size of this meta-analysis statistically not signif-
icant ( p⬎.05). The fail-safe value (Rosenthal, 1979) for this
study indicates that there would have to be over 1,000 such hidden
studies with outcome-alliance relation r⫽.0 before the aggregate
ES for the overall alliance-outcome relation would cease to be
statistically significant. (A highly unlikely scenario).
Another way to explore the possibility of a systematic bias in the
data is by inspecting the funnel plot. A funnel plot is a diagram of
standard error on the vertical axis as a function of effect size on the
horizontal axis. In the presence of bias, one would expect the plot
to show a higher concentration of studies on one side of the mean
than the other. Typically, smaller sample size studies (having
larger standard errors) are more likely to be published if they have
larger than average effects. In the absence of publication bias we
would expect the studies to be distributed relatively symmetrical
around the aggregated ES. The plot of our data did not indicate a
strong bias.
Two other possible sources of systematic trends were investi-
gated: 1) Date of publication: Did the study ES, on the average,
change in magnitude over time? and 2) Study sample size: Were
the effect sizes related to the number of participants in the study?
There was a small and statistically nonsignificant negative time-
trend observed ( p⫽.082). Over time (1972–2009) researchers
were reporting slightly decreasing ESs. This makes intuitive sense
because recent studies use more sophisticated methods for con-
trolling for pre and in-therapy effects such as problem severity,
early symptom reduction, and so forth. These variables share some
of the variance with the relation between alliance and outcome.
Statistically controlling for these factors would remove variance in
outcome that earlier investigators did not partial out. Thus, using
these procedures likely exerts a downward pressure on the corre-
lation between alliance and outcome.
More surprisingly, we found a significant relation between
sample size and ES (r⫽⫺.25, p⬍.01). The best fitting regression
line for this puzzling association is nonlinear; the studies with
sample sizes between 100 and 200 appear to report lower ESs
compared to studies with both smaller and larger sample sizes.
This effect may be an artifact of some sort, but will require further
investigation.
In sum, the overall relation between alliance and outcome in
individual psychotherapy is robust, not effected by the file drawer
problem, and accounts for approximately 7.5% of the variance in
treatment outcomes.
Variability of Effect Sizes
Similar to what was found in the previous meta-analysis (Hor-
vath & Bedi, 2002), the alliance-outcome relations in this data set
were not homogenous (Q ⫽498.42, df 189, p⬍.00001). We
computed the I
2
statistic which provides an estimate of the per-
centage of variance of ESs over and above the amount of variabil-
ity that can be accounted for by random (chance) variation. The I
2
of .56 we obtained indicates that the variability in alliance outcome
correlations among the collection of research we gathered is ap-
proximately 56% greater than one would expect if all the studies
were measuring the same concept. This finding, in itself, is not
surprising. The research we were analyzing covered a wide range
of treatments, a broad cross section of client problems, we had
process (alliance) and outcome measures that came from different
sources (therapist, client, or observers), and outcomes were mea-
sured from a variety of perspectives, at different times; sometimes
immediately after treatments, at other times at follow-up points.
Each of these factors could act as a moderator of the relation
between alliance and outcome and explain the excess variability
among the ES. We then proceeded to investigate the effect of some
of these potential moderators.
Moderator Analyses
We investigated the impact of six categorical variables that have
the potential of moderating the relation between alliance and
outcome: alliance measure (CALPAS, VPPS, HAq, WAI, and
Other); alliance rater (client, therapist, observer); time of alliance
assessment (Early, Mid, Late, Averaged); outcome measure (BDI,
SCL, Dropout); type of treatment (CBT, IPT, Psychodynamic,
Substance Abuse); and publication source (journal, books/
chapters, unpublished/thesis). Table 1 provides a summary of the
results of these analyses. Each of the analyses were done using
independent data; when a researcher provided more than one result
related to the moderator only one ES, randomly chosen, was used
in the contrast. The only exception to this rule was in the case of
“time of alliance assessment.” In this analysis, the data are inde-
pendent, but rather than choosing randomly among the available
ES, we choose to include early alliance measure, if it was avail-
able, because of the clinical importance of this level of the vari-
able.
There are several noteworthy features that apply to all of these
results: All of the aggregate alliance-outcome correlations in each
category are statistically significant beyond p⬍.001. This result
strongly supports the claim the impact of the alliance on therapy
outcome is ubiquitous irrespective of how the alliance is measured,
from whose perspective it is evaluated, when it is assessed, the
way the outcome is evaluated, and the type of therapy involved.
The quality of the alliance matters.
The next most common feature is the finding that, with very few
exceptions, within each of these subsets of data, the ES are very
diverse in magnitude. We noted earlier that heterogeneity of the
ESs in a large-scale meta-analysis is not unusual. However, these
results indicate that the high degree of variability remains practi-
cally unchanged within each level of these potential moderators.
One of the consequences of such large variances within the levels of
these variables was, that while the aggregate ES between the levels of
these potential moderators did show some of the patterns we would
13
SPECIAL ISSUE: ALLIANCE IN INDIVIDUAL PSYCHOTHERAPY

expect (e.g., the closer the alliance was assessed to termination the
higher the correlation between alliance and outcome became, etc.)
only the contrast between the outcome measures BDI and dropout
was statistically reliable. All the other differences failed to reach
statistical significance due the high levels of variances within each
category of moderators.
The Halo Effect
We revisited a question raised in earlier meta-analyses: Are the
alliance-outcome relations researchers report inflated by a “halo
effect”? That is, did the common practice of sourcing both process
(alliance) and outcome data from the same individual artificially
inflate the magnitude of the relation between alliance and out-
come? To evaluate this question we contrasted the aggregate ES
derived from studies using the same source (raters) with the
aggregate ES derived from studies where the alliance and outcome
data came from different sources. The aggregate ES for alliance-
outcome correlations generated by the same source for both rating
is r⫽.29 (based on 94 ES), the value for the ES based on different
raters is r⫽.25 (based on 96 ES). The difference between these
values is not statistically significant Q
b
⫽3.09 df ⫽1, p⫽.078.
However it may be too early to close the door on this question; the
heterogeneity within each of these categories was 54% greater than
expected by chance. As noted before, such excess of variance
significantly dampens the possibility of finding statistically signif-
icant differences in contrasts. In addition, there may be a trend over
time; the ES from alliance outcome relations derived from different
sources has been less than the same source correlation each time this
question was examined, and with each subsequent test of this hypoth-
esis, the gap between these values has crept closer to statistically
significant levels (Horvath & Bedi, 2002; Horvath & Symonds,
1991).
Limitations of the Research
This report is based on a numerical synthesis of all the research
results currently available that met our inclusion criteria. By in-
cluding all research in which the authors refer to the process
variable as alliance, we might have collected and summarized a
number of different kinds of things. This is a serious concern,
especially in light of the fact that the ESs in this data set are quite
diverse. A practical response to this conceptual problem is to
conclude that this meta-analysis reports the results of alliance-
outcome relation as it is researched at this time.
There are also some technical constrains effecting these analy-
ses. We chose to use independent data whenever possible. To
achieve this, on many occasions we needed to randomly discard
some data (ES) in order to make sure that only one result from a
particular research report was used in each analysis. As a result, we
lost some power to detect differences in a number of analyses. In
the long run, the use of independent data is statistically well
justified, but the resulting constraints on the computations are also
important for the reader to consider.
Table 1
Categorical Moderators of the Alliance-Outcome Relation (Based on Independent Samples)
Moderator Categories (k) ES (r)pwithin category I
2
pbetween
categories
Alliance measure Other (57) .27 p⬍.000 .59 NS
CALPAS (28) .23 p⬍.000 .25
HAq (31) .29 p⬍.000 .75
VPPS (5) .29 p⬍.000 .0
WAI (80) .28 p⬍.000 .51
Alliance rater Client (112) .28 p⬍.000 .64 NS
Therapist (23) .20 p⬍.000 .57
Observer (40) .29 p⬍.000 .04
Time of assessment Early (113) .25 p⬍.000 .52 p⬍.001
Mid (33) .25 p⬍.000 .39
Late (36) .39 p⬍.000 .71
Averaged (52) .31 p⬍.000 .56
Outcome Measure
ⴱ
BDI (27) .42 p⬍.000 .56 p⬍.001
SCL (44) .27 p⬍.000 .55
ⴱ
Dropout (19) .18 p⬍.000 .14
Type of treatment CBT (28) .35 p⬍.000 .64 NS
IPT (9) .39 p⬍.000 .39
Psychodynamic (44) .29 p⬍.000 .57
Substance Abuse (17) .23 p⬍.000 .80
Publication source Journal (141) .29 p⬍.000 .61 NS
Unpublished (Thesis) (21) .28 p⬍.000 .55
Chapter (16) .22 p⬍.000 .43
ⴱ
Post hoc contrast (Tukey) of difference between these values is significant p⬍.05.
14 HORVATH, DEL RE, FLU
¨CKIGER, AND SYMONDS

Therapeutic Practices
The positive relation between the quality of the alliance and
diverse outcomes for many different types of psychological ther-
apies is confirmed in this meta-analysis. While the overall ES of
r⫽.275 accounts for a relatively modest proportion of the total
variance in treatment outcome, the magnitude of this correlation,
along with therapist effects, is one of the strongest and most robust
predictors of treatment success empirical research has been able to
document (Wampold, 2001). The practice recommendations pre-
sented below are based on important studies of the alliance, but not
necessarily grounded in the kinds of statistical procedures we used
in our meta-analysis:
}The development and fostering of the alliance is not separate
from the interventions therapist implement to help their clients; it
is influenced by and is an essential and inseparable part of every-
thing that happens in therapy. In this sense, the therapist does not
“build alliance” but rather he or she does the work of treatment in
such a way that the process forges an alliance with the client. The
quality of the alliance is an index of the level of mutual and
collaborative commitment to the “business of therapy” by therapist
and client. Its distinguishing feature is the focus on therapy as a
collaborative enterprise. Another way to grasp the concept is that
it is a measure of how well the therapist and client work together.
}The development of a “good enough” alliance early in therapy
is vital for therapy success: On one hand, establishing a good
alliance prevents clients dropping out and, on the other hand, the
sense of collaboration creates a “working space,” with room to
introduce new ways of addressing the clients concerns.
}In the early phases of therapy, modulating the methods of
therapy (tasks) to suit the specific client’s needs, expectations and
capacities is important in building the alliance. The therapist and
client need to find the level of collaboration suited to achieve the
work of therapy taking into account not only the clients’ problems,
but also the resources, capacities, and expectations they bring to
therapy. Bridging the client’s expectations and personal resources
and what the therapist believes to be the most appropriate inter-
vention is an important and delicate task. Alliance emerges, in part,
as a result of the smooth coordination of these elements.
}Therapist and client perceptions of the alliance, particularly early
in treatment, do not necessarily match. Misjudging the client’s felt
experience of the alliance (i.e., believing that it is in “good shape”
when the client does not share this perception) could render therapeu-
tic interventions less effective. Active monitoring the clients’ alliance
throughout treatment is a recommended practice.
}The strength of the alliance, within or between sessions, often
fluctuates in response to a variety of in-therapy factors, such as
therapists challenging clients to grapple with difficult issues, mis-
understandings, transference, and so forth. These “normal” varia-
tions—as long as they are attended to and resolved—are associated
with good treatment outcomes.
}Therapists’ nondefensive responses to client negativity or
hostility are critical for maintaining a good alliance. Therapists
have to develop the ability to neither internalize nor to ignore
clients’ negative responses.
}Recent studies suggest that therapists’ contributions to the
quality of the alliance are critical. Therapists who are good at
building a strong alliance tend to have better alliances with most of
their clients. The reverse is also true. This finding suggests that
alliance development is a skill and/or capacity that therapist can
and should be trained to develop just as they are trained to attend
to other aspects of their practice.
References
A reference list of the 201 studies in used in the meta analysis is available
from the first author and will be published in the chapter by the same
authors: “Alliance in Individual Therapy” in J. C. Norcross (Ed.), 2011,
Psychotherapy relationships that work (2nd ed.). New York: Oxford
University Press.
Bordin, E. S. (1975, September). The working alliance: Basis for a general
theory of psychotherapy. Paper presented at the Society for Psychother-
apy Research, Washington, DC.
Bordin, E. S. (1980, June). Of human bonds that bind or free. Paper
presented at the Society for Psychotherapy Research, Pacific Grove, CA.
Bordin, E. S. (1989, June). Building therapeutic alliances: The base for
integration. Paper presented at the Society for Psychotherapy Research,
Berkley, CA.
Bordin, E. S. (1994). Theory and research on the therapeutic working
alliance: New directions. In A. O. Horvath & L. S. Greenberg (Eds.), The
working alliance: Theory, research, and practice (pp. 13–37). New
York: Wiley.
Cooper, H., Hedges, L. V., & Valentine, J. C. (2009). The handbook of
research synthesis and meta-analysis (2nd ed.). New York: Russell Sage
Foundation.
Del Re. A. C. (2010). RcmdrPlugin. MAc: Meta-Analysis with Correla-
tions (MAc) Rcmdr Plug-in. R package version 1.0.5. [Computer soft-
ware] Available from http://CRAN.Rproject.org/package⫽
RcmdrPlugin.MAc
Del Re. A. C., & Hoyt, W. T. (2010). MAc: Meta-analysis with correla-
tions. R package version 1.0.5. [Computer software] Available from
http://CRAN.R-project.org/package⫽MAc
Fisher, R. A. (1924). On a distribution yielding the error functions of
several well known statistics. Proceedings of the International Congress
of Mathematics, Toronto, 2, 805– 813.
Frank, J. D., & Frank, J. B. (1991). Persuasion and healing: A comparative
study of psychotherapy (3rd ed.). Baltimore, MD: Johns Hopkins Uni-
versity Press.
Freud, S. (1912/1958). The dynamics of transference [Zur Dynamik der
U
¨bertragung]. (J. Starchey, Trans.). In J. Starchey (Ed.), The standard
edition of the complete psychological works of Sigmund Freud (vol. 12,
pp. 99 –108). London: Hogarth Press.
Freud, S. (1913). On the beginning of treatment: Further recommendations
on the technique of psychoanalysis [Zur Einleitung der Behandlung –
Weitere Ratschla¨ge zur Technik der Psychoanalyse]. In J. Strachey
(Ed.), Standard edition of the complete psychological works of Sigmund
Freud (vol. 12, pp. 122–144). London: Hogarth.
Gaston, L., Goldfried, M. R., Greenberg, L. S., Horvath, A. O., Raue, P. J.,
& Watson, J. (1995). The therapeutic alliance in psychodynamic,
cognitive-behavioral and experiential therapies. Journal of Psychother-
apy Integration, 15, 1–26.
Goldfried, M. R. (1980/2009). Toward the delineation of therapeutic
change principles. Applied and Preventive Psychology, 13, 3– 4.
Greenson, R. R. (1965). The working alliance and the transference neuro-
ses. Psychoanalysis Quarterly, 34, 155–181.
Hatcher, R. L., & Barends, A. W. (1996). Patients’ view of the alliance of
psychotherapy: Exploratory factor analysis of three alliance measures.
Journal of Consulting and Clinical Psychology, 64, 1326 –1336.
Hatcher, R. L., & Barends, A. W. (2006). Thinking about the alliance in
practice. Psychotherapy, Therapy, Research, Practice, Training, 41,
7–10.
Hatcher, R. L., Barends, A., Hansell, J., & Gutfreund, M. J. (1995).
Patient’s and therapist’s shared and unique views of the therapeutic
15
SPECIAL ISSUE: ALLIANCE IN INDIVIDUAL PSYCHOTHERAPY

alliance: An investigation using confirmatory factor analysis in a nested
design. Psychoanalysis Quarterly, 63, 636 – 643.
Horvath, A. O. (2009, October). Conceptual and methodological chal-
lenges in alliance research Is it time for a change? Paper presented at the
Society for Psychotherapy Research European Division, Bolzano, Italy.
Horvath, A. O., & Bedi, R. P. (2002). The alliance. In J. C. Norcross (Ed.),
Psychotherapy relationships that work. New York: Oxford University
Press.
Horvath, A. O., Del Re, A., Flu¨ ckiger, C., & Symonds, D. B. (2011). Alliance
in individual psychotherapy. In J. C. Norcross (Ed.), Psychotherapy rela-
tionships that work (2nd. ed.). New York: Oxford University Press.
Horvath, A. O., Gaston, L., & Luborsky, L. (1993). The role of alliance in
psychotherapy. In N. Miller, L. Luborsky, J. Barber, & J. P. Docherty
(Eds.), Psychodynamic treatment research: A handbook for clinical
practice (pp. 247–274). New York: Basic Books.
Horvath, A. O., & Luborsky, L. (1993). The role of the therapeutic alliance
in psychotherapy. Journal of Consulting and Clinical Psychology, 61,
561–573.
Horvath, A. O., & Symonds, B. D. (1991). Relation between working
alliance and outcome in psychotherapy: A meta-analysis. Journal of
Counseling Psychology, 38, 139 –149.
Hunter, J. E., & Schmidt, F. L. (2004). Methods of meta-analysis (2nd ed.).
Thousand Oaks, CA: Sage.
Luborsky, L. (1976). Helping alliances in psychotherapy. In J. L. Clegh-
horn (Ed.), Successful psychotherapy (pp. 92–116). New York: Brunner/
Mazel.
Luborsky, L., Singer, B., & Luborsky, L. (1975). Comparative studies of
psychotherapies; “Is it true that everybody has won and all must have
prizes”? Archives of General Psychiatry, 32, 995–1008.
Marmar, C. R., Horowitz, M. J., Weiss, D. S., & Marziali, E. (1986). The
development of the therapeutic alliance rating system. In L. S. Green-
berg & W. M. Pinsof (Eds.), The psychotherapeutic process: A research
handbook (pp. 367–390). New York: Guilford Press.
Martin, D. J., Garske, J. P., & Davis, K. M. (2000). Relation of the
Therapeutic alliance with outcome and other variables: A meta analytic
review. Journal of Clinical and Consulting Psychology, 68, 438 – 450.
National Research Council. (1992). Combining information: Statistical
issues and opportunities for research. Washington DC: National Acad-
emy Press.
R Development Core Team. (2009). R: A language and environment for
statistical computing. [Computer software]. R Foundation for Statistical
Computing, Vienna. ISBN 3–900051-07– 0. Retrieved from http://
www.R-project.org
Rogers, C. R., & Wood, J. K. (1974). Client-centered theory: Carl R.
Rogers. In A. Burton (Ed.), Operational theories of personality (pp.
211–258.). New York: Bruner/Mazel.
Rosenthal, R. (1979). The file drawer problem and tolerance for null
results. Psychological Bulletin, 86, 638 – 641.
Rosenzweig, S. (1936). Some implicit common factors in diverse methods
of psychotherapy. American Journal of Orthopsychiatry, 6, 412– 415.
Smith, M. L., & Glass, G. V. (1977). Meta-analysis of psychotherapy
outcome studies. American Psychologist, 32, 752–760.
Sterba, R. F. (1934). The fate of the ego in analytic therapy. International
Journal of Psychoanalysis, 115, 117–126.
Stiles, W. B., Shapiro, D., & Elliot, R. (1986). Are all psychotherapies
equivalent? American Psychologist, 41, 165–180.
Stone, L. (1961). The psychoanalytic situation: An examination of its
development and essential nature. New York: International Universities
Press.
Sutton, A. J. (2009). Publication bias. In H. Cooper, L. V. Hedges, & J. C.
Valentine (Eds.). The handbook of research synthesis and meta-analysis
(2nd ed.). New York: Russell Sage Foundation.
Viechtbauer, W. (2005). Bias and efficiency of meta-analytic estimators in
the random-effects model. Journal of Education and Behavioral Statis-
tics, 30, 261–293.
Wampold, B. E. (2001). The great psychotherapy debate: Models, meth-
ods, and findings. Mahwah: Erlbaum.
Wampold, B. E., Mondin, G. W., Moody, M., Stich, F., Benson, K., &
Ahn, H. (1997). A meta-analysis of outcome studies comparing bona
fide psychotherapies: Empirically, “all must have prizes.” Psychological
Bulletin, 122, 203–215.
Zetzel, E. R. (1956). Current concepts of transference. International Jour-
nal of Psychoanalysis, 37, 369 –376.
Received August 9, 2010
Revision received September 27, 2010
Accepted September 27, 2010 䡲
E-Mail Notification of Your Latest Issue Online!
Would you like to know when the next issue of your favorite APA journal will be available
online? This service is now available to you. Sign up at http://notify.apa.org/ and you will be
notified by e-mail when issues of interest to you become available!
16 HORVATH, DEL RE, FLU
¨CKIGER, AND SYMONDS