Psychotherapy © 2011 American Psychological Association

2011, Vol. 48, No. 1, 17–24 0033-3204/11/$12.00 DOI: 10.1037/a0022181

The Alliance in Child and Adolescent Psychotherapy

Stephen R. Shirk Marc S. Karver and Renee Brown

University of Denver University of South Florida

The therapeutic alliance has a long history in the child and adolescent psychotherapy literature. This
article examines prominent views on the alliance with youth and considers a number of issues that
distinguish youth alliance from its adult counterpart. A meta-analysis of alliance– outcome associations
in individual youth therapy is presented. In order to provide a direct comparison with the adult literature,
the review included only prospective studies of individual youth therapy that used an explicit measure of
alliance. Results from 16 studies revealed consistency with the adult literature with a weighted mean
correlation of .22 (k ⫽ 16, n ⫽ 1306, p ⬍ .001) between alliance and outcome (CI ⫽ ⫹/⫺.06). Although
there were trends showing stronger alliance– outcome associations for child versus adolescent therapy
and for behavioral versus nonbehavioral therapies, only problem type (substance abuse and mixed
problems vs. eating disorders) significantly moderated alliance– outcome associations. Limitations of the
research and implications for therapeutic practice are discussed.

Keywords: alliance, child, adolescent, psychotherapy, meta-analysis

The therapeutic alliance has a long history in the child and tional relationship and the collaborative relationship (Estrada &
adolescent psychotherapy literature dating to the work of Anna Russell, 1999; Shirk & Saiz, 1992). Of equal importance, the link
Freud (1946). In contrast, research on the alliance in youth therapy between bond and collaboration is framed functionally; the emo-
is relatively new. Not long ago, Shirk and Karver (2003) identified tional bond enables the child to work purposefully on the tasks of
only one study that met inclusion criteria used in adult alliance therapy. The bond itself is not posited as curative, but rather as a
meta-analyses. Since that review, additional youth alliance studies catalyst for promoting therapeutic work. It is interesting that this
have emerged though the total number of studies still pales in view is revived in later cognitive– behavioral formulations of the
comparison to the adult literature. therapy relationship. The alliance serves specific technical proce-
Our primary aim in this article is to provide a current assessment dures and can facilitate child involvement in therapy tasks ranging
of alliance– outcome relations in individual child and adolescent
from in-session exposures to between session homework comple-
therapy through a meta-analysis of existing studies. We preface
tion (Kendall, Comer, Marker, Creed, & Puliafico, 2009; Shirk &
this analysis with an examination of the alliance construct in youth
Karver, 2006).
therapy and consider developmental issues that distinguish adult
In contrast to this perspective, play therapists have long empha-
and youth alliance. Finally, limitations of the research and thera-
peutic practices are presented. sized the curative nature of the therapy relationship (Axline, 1947).
In this tradition, the child’s experience of the therapist as support-
ive, attuned, and nonjudgmental is pivotal for therapeutic change
Definitions and Measures
(Shirk & Russell, 1996; Wright, Everett, & Roisman, 1986). Draw-
Two views of the therapeutic relationship were prominent in the ing on the work of Rogers (1957), therapy was not conceptualized
early history of child therapy. Anna Freud (1946) observed that an as treatment, as something you do to the child, but rather as an
“affectionate attachment” between child and therapist is a “pre- opportunity for growth. The relational conditions of empathy,
requisite for all later work” in child therapy (p. 31). In this early genuineness, and positive regard were posited as the active ingre-
statement, we find an enduring differentiation of alliance compo- dients of therapy. Associations between bond and outcome in this
nents: the distinction between bond and work, between the emo- tradition are direct rather than mediated through therapeutic work.
Common to the foregoing perspectives is an emphasis on an
emotional connection between child and therapist. Emotional
Stephen R. Shirk, Department of Psychology, University of Denver; bond, then, appears to be a core component of alliance with
Marc S. Karver and Renee Brown, Department of Psychology, University children. This view has taken root in recent approaches to assess-
of South Florida. ing the alliance in child and adolescent therapy (e.g., Shirk & Saiz,
This article is adapted, by special permission of Oxford University 1992; Shirk & Russell, 1996; Shirk, Gudmundsen, Kaplinski, &
Press, from a chap. of the same title by the same authors in J. C. Norcross McMakin, 2008). Some have criticized this perspective for failing
(Ed.), 2011, Psychotherapy relationships that work (2nd ed.). New York:
to acknowledge the social contractual features of the therapeutic
Oxford University Press. The book project was cosponsored by the APA
Division of Psychotherapy.
alliance (DiGiuseppe, Linscott, & Jilton, 1996). From this perspec-
Correspondence concerning this article should be addressed to Stephen tive, a central component of alliance, especially with older children
R. Shirk, Department of Psychology, University of Denver, Denver, CO and adolescents, consists of agreements regarding treatment goals
80208. E-mail: sshirk@du.edu and the methods for accomplishing them. The fact that youth are
17
18 SHIRK, KARVER, AND BROWN

typically referred by others makes the establishment of agreements parents are more actively involved as informants about client
both difficult and essential for treatment collaboration. functioning, collateral participants, or even as therapeutic collab-
As these clinical perspectives suggest, there are important par- orators who help with treatment implementation outside sessions.
allels between adult and youth models of alliance. Consistent with Consequently, the therapist is faced with multiple sets of goals,
Bordin’s (1979) pan theoretical model, three facets of alliance— and often the goals of parents and youth diverge. Agreement on
emotional bond, task collaboration (work), and agreements (goal goals, then, is complicated by whose goals are considered. A study
consensus)—are prominent in the youth literature. Although it is of clinic-referred children (Hawley & Weisz, 2003) examined
tempting to view this convergence as evidence for configural therapist, child, and parent agreement about the most important
invariance across age groups, at least two studies have failed to problems to be addressed in therapy. Amazingly, more than 75%
fully support the three factor model with youth. These studies of child, parent, and therapist triads began treatment without
produced a single factor solution, thus suggesting that features of agreement on even one target problem. Nearly half failed to agree
the alliance may be less differentiated at younger ages (DiGiuseppe et on one broad problem domain such as aggression versus depres-
al., 1996; Faw, Hogue, Johnson, Diamond, & Liddle, 2005). Oth- sion. It is interesting to note that therapists agreed with parents
ers have shown that emotional bond and task collaboration repre- more often than with children. Such evidence suggests that agree-
sent distinct but correlated alliance dimensions (Estrada & Russell, ment on goals may mean something quite different in youth
1999; Shirk & Saiz, 1992). therapy than in adult therapy. At present, it is not clear if agree-
ment between parent and therapist or child and therapist is a better
Developmental Considerations predictor of treatment outcome.
A related issue involves the presence of multiple alliances in
Across the dimensions of bond, task, and goals, youth and adult youth treatment even when it is child focused. Unlike with adults,
alliance are distinguished by a number of developmental factors. therapists are faced with establishing and maintaining an alliance
Consider first the therapy bond. Anna Freud (1946) noted that a with the youth and his or her parent(s). Most research on alliance–
child’s relationship with the therapist could arise from a number of outcome relations with children and adolescents has focused on the
sources, not all of them developmentally equivalent. For many youth–therapist relationship. A notable exception is in the area of
children, the relationship with a therapist is an opportunity to fulfill parent management training where parents are the primary focus of
needs not available in other contexts. As A. Freud observed, “if no child therapy. It is possible that alliances with parents and youth
one at home plays games with the child, for example, he might like relate to different sets of outcomes. For example, Hawley and
to come to treatment because there a grown-up pays attention to Weisz (2005) found that parent, but not youth, alliance predicted
him” (Sandler, Kennedy, & Tyson, 1980, p. 47). Children’s pos- better therapy participation. Youth alliance, but not parent alliance,
itive feelings for the therapist may be connected to features not predicted symptom change. These findings suggest that a strong
typically regarded as therapeutic, for example, how fun, stimulat- alliance with parents is important for treatment continuation,
ing, or rewarding the therapist might be. In such cases it is unclear whereas the youth alliance may be more critical for treatment
if the “bond” reflects an experience of the therapist as an “ally,” or outcomes.
as a valued playmate. In fact, A. Freud (1946) distinguished this
type of relationship from the alliance. In the latter, the therapeutic Youth Alliance Measurement
bond is based on experiencing the therapist as someone who can be
counted on for help with emotional or behavioral problems. Al- Despite its relatively small size, our review of research on the
though the bond may be closely linked with this function in adult youth alliance uncovered 10 different alliance measures for chil-
therapy, it cannot be assumed with youth. dren and adolescents. No study has examined the concurrent
A second developmental issue concerns the task dimension of validity of the most frequently used measures of youth alliance;
alliance. In the adult literature, tasks are framed in terms of consequently, it is not clear if different measures with similar
agreements about the content and methods of therapy; in essence, names are assessing the same construct.
whether there is consensus between client and therapist on the The two most frequently used patient and therapist report in-
substance of therapy (Bordin, 1979). Such judgments may exceed struments in the youth literature are the Working Alliance Inven-
the cognitive capacities of many child and adolescent clients. For tory (WAI; Horvath & Greenberg, 1989) and the Therapeutic
example, studies of children’s understanding of therapy have Alliance Scale for Children (TASC; Shirk & Saiz, 1992). The
shown important developmental progressions in their recognition WAI has been used primarily with adolescents and the TASC with
of therapy processes and parameters (e.g., Shirk & Russell, 1998, children and young adolescents. Although the WAI, originally
for a review). For example, children’s causal reasoning may limit developed for adult therapy, has been modified for use with
their ability to understand links between specific therapy tasks and adolescents (Linscott, DiGuiseppe & Jilton, 1993), the original or
subsequent therapy goals (Shirk, 1988). Such developmental con- short version has been employed most frequently. The WAI mea-
cerns have prompted some investigators to suggest that task col- sures the quality of the therapeutic relationship across three sub-
laboration with children is best assessed through observation rather scales: bonds, tasks, and goals. The final item pool for the measure
than self-report (Karver et al., 2008; Shirk & Karver, 2006). was generated on the basis of content analysis of Bordin’s (1979)
Developmental issues complicate the goal dimension of the model of working alliance. Exert raters evaluated items for good-
alliance as well. An important difference between adult and youth ness of fit with the working alliance construct.
therapy is the involvement of other family members aside from the The TASC was developed specifically for child therapy and also
identified client. Minimally, parents or guardians are involved in was based on Bordin’s model (1979). Two dimensions are as-
transportation to and payment of therapy. Quite often, however, sessed: bond between child and therapist, and level of task collab-
 SPECIAL ISSUE: ALLIANCE IN YOUTH THERAPY 19

oration. Unlike the WAI, task collaboration does not refer to In the next example, the goal dimension of the alliance is
agreements on tasks, but to ratings of actual collaboration on tasks prominent. Here the therapist explores the adolescent’s goals for
such as “talking about feelings” and “trying to solve problems.” therapy.
The therapist version of the TASC involves ratings of the child’s
bond and task involvement rather than the therapist’s own. Al- Therapist: I know what your parents are hoping for from our
though items on the bond subscale remain constant, items on the therapy, but what are your goals?
task collaboration scale can be modified to fit specific types of
Client: I want to stop worrying that I’ll say something wrong,
therapy, though to this point, only the original set of items have
so I won’t just feel all stuck.
been used in research. The subscales show good internal consis-
tency and relatively high levels of stability over a 4- to 7-week Therapist: It sounds like you’d like to shift your focus away
period (Shirk et al., 2008). from all the things that make you worry and feel trapped.
A number of observational measures have appeared in the youth
literature but none has become the “gold standard.” One measure Client: Yeah, when I think about what could happen, I be-
that was developed specifically for child and adolescent therapy come so nervous I just avoid everyone. I want to go walk up
and that has been used in more than one study is the Therapy to the ins (popular girls) and just be right there talking and not
Process Observation Coding System – Alliance Scale (McLeod & all what if.
Weisz, 2005). This observational scale took as its starting point the
Therapist: So, if we could change how much you worry and
distinction between bond and task collaboration found in factor
think about all the negatives that would be a good result?
analyses of child and therapist reports of alliance (Shirk & Saiz,
1992) and factor analyses of process codes (Estrada & Russell, Client: Definitely, I’m tired of worrying all the time.
1999). Items from a broad range of measures that mapped onto the
bond and task dimensions were initially included in the item pool, In the final example, an older child talks with his therapist about
and redundant items reduced. Expert raters then sorted items into dealing with anger. The client’s statements reflect the collaborative
bond or task categories and consistently sorted items were re- aspect of the alliance:
tained. The resulting coding system includes eight bond items and
six task collaboration items. Interrater reliability has been shown to Client: I feel better since we last talked. That stuff we worked
be good across items. Bond and task dimensions are highly cor- on was pretty helpful.
related, consistent with what has been found with youth self-
Therapist: That’s cool. Great. What did you do?
reports of alliance dimensions, suggesting that alliance may be a
unitary construct in youth therapy. Client: Like. . . I forgot what it is called. . . like. . . I con-
trolled my temper. . . when I got angry. . . I was like ok like
take a deep breath. . . then I walked away.
Clinical Examples
Therapist: Great. It helped bring your anger down.
The following interactions derived from a composite of cases
reflect different features of the therapeutic alliance with young Client: Mmhmm.
clients. The first example illustrates a strong emotional bond
between a young adolescent and her therapist: Therapist: You made a good decision. Some people get angry
and are like, hey, I’m right, I’m not backing down.
Therapist: So, what is it like when you’re feeling really
down? Client: If I get up in their face when I’m mad, I end up losing
anyway.
Client: I get like I don’t want to talk to anyone. I’m like get
away, leave me alone. My Dad asks me how I’m doing and I Therapist: Losing anyway?
just say nothing or walk away.
Client: Yeah, I pay for it later. Get in trouble and stuff.
Therapist: You just want some space. You don’t want to be
pushed. Therapist: So using what we’ve worked on might have a pay
off?
Client: Exactly.
Client: Yeah, like what we practiced in here.
Therapist: In here, I’m going to ask you a lot about how you
are feeling. If you feel like I’m pushing you, is it possible you Meta-Analytic Review
will not want to talk with me.
Consistent with the adult alliance literature, previous meta-
Client: I don’t think that’ll happen because you’re not in my analyses have revealed small to moderate associations between
face. Talking gets my stress out. When I’m in a bad mood on alliance and outcome in youth therapy (Shirk & Karver, 2003;
the day of our meetings, I look forward to our talking. . . it Karver Handelsman, Fields, & Bickman, 2006). However, conclu-
helps keep me going because I know you get me. sions from prior reviews have been constrained by the limited
20 SHIRK, KARVER, AND BROWN

number of studies that actually included a bone fide alliance The resulting sample consisted of 16 studies with 658 youth
measure, rather than a general relationship measure, and by the clients (208 children, 395 adolescents, and 55 unclassified youth)
inclusion of studies that assessed alliance and outcome concur- and 648 parents. Studies were also coded for age of client (child,
rently rather than prospectively. In addition, prior meta-analyses adolescent, or parent), type of treatment (behavioral vs. nonbehav-
have combined results from diverse forms of therapy, for example, ioral) and problem type following procedures reported by Shirk
individual, group, milieu, and family therapies, without adequate and Karver, 2003).
sample size to evaluate each form separately. Our comprehensive
review of published and unpublished studies on alliance– outcome
relations revealed that there are now enough studies in the youth Calculation of Effect Sizes
literature to sharpen our focus on prospective studies of individual
Because most studies reported results (alliance to outcome re-
child and adolescent therapy (Shirk & Karver, in press).
lationships) as correlations, the product-moment correlation coef-
In order to overcome the limitations of previous meta-analyses,
ficient r was used as our effect size estimate. All results in each
we examined alliance– outcome associations in studies of child and
study were converted to Fisher’s Z in order to normalize the r
adolescent therapy that included an explicit alliance measure and
distribution (Hedges & Olkin, 1985). For all studies it was possible
evaluated associations prospectively. In order to make a direct
to compute effect sizes, thus no effects were imputed as zero. In
comparison with the adult literature, we also limited our review to
most studies, more than one alliance– outcome relationship was
individual therapy. Thus, our meta-analysis does not include stud-
reported. In order to correct for bias due to correlated effects
ies of family, group, or milieu therapy. Because parents are often
within studies and an unequal number of associations reported in
included in child and adolescent therapy, in fact, they may be the
different studies, we averaged (simple mean) the Fisher’s Z’s for
focus of behavioral treatments, we also provide an estimate of
each study. In order to calculate a more precise estimate of the
association between parent alliance and outcome.
overall relation between alliance and outcome utilizing a fixed
Our analyses addressed four questions. First, what is the pro-
effects model (Hedges, 1994), we weighted the average effect
spective association between alliance and outcome in individual
size (Z) for each study by the number of participants in the study.
youth therapy? Second, are there age-related differences in
We weighted each effect size so that the final estimate of the
alliance-outcome associations across children, adolescents, and
alliance to outcome relationship properly accounted for the fact
parents? Third, does strength of alliance– outcome associations
that more precise estimates should be given more weight in the
vary across broad treatment types, that is, behavioral versus non-
aggregate. The weighted effect sizes for each study were aggre-
behavioral therapy? And fourth, are alliance– outcome associations
gated and then this sum was divided by the sum of the weights
moderated by type of treated problem?
(number of participants per study minus three) for each study,
resulting in an estimate of the overall alliance to outcome relation (wi
Selection of Studies * ESi/Sum wi; wi ⫽ n – 3; Hedges & Olkin, 1985), This weighted
effect size Z was then converted back to the product-moment
To identify applicable studies, a three pronged approach was
correlation coefficient r. Because of small sample size, we com-
used. First, prior reviews of the alliance to outcome relationship
puted 95% confidence intervals for subsets of correlations, for
were examined for qualifying manuscripts (Shirk & Karver, 2003;
example, behavioral versus nonbehavioral, in order to determine
Karver et al., 2006). Citations of these articles were then examined
significant differences across groups.
as a means to identify additional manuscripts. Second, the
PsycINFO database was searched from 2004 forward to identify
articles that have been published since the last major meta-analytic Results of the Meta-Analysis
review of the therapeutic alliance in child and adolescent therapy.
Finally, Google Scholar was used to search for studies that may The 16 studies that met inclusion criteria are displayed in Table 1.
have been missed and for unpublished manuscripts. For both Together the studies included 1306 participants with 658 youth
searches, child and adolescent were used in conjunction with the clients (208 children, 395 adolescents, and 55 unclassified youth)
terms alliance or relationship and therapeutic or therapy. and 648 parents. The sample includes 13 published studies and
To be included in the current meta-analysis, studies had to meet three doctoral dissertations. Youth alliance was assessed in 14
the following criteria: (1) include a specific measure explicitly studies (four child, nine adolescent, and one combined age group)
described in the manuscript as an alliance measure; (2) the alliance and parent alliance in six. Eight studies included a behavioral or
had to be related to some indicator or measure of post treatment cognitive– behavioral treatment and nine included a nonbehavioral
outcome and not another process variable, including sessions com- therapy. Of the 16 studies, five included internalizing samples,
pleted; (3) the study had to be of individual therapy delivered to a four included mixed problem samples, three included externalizing
youth under age 18 or a parent; (4) alliance needed to be measured samples, three included substance abuse samples, and one in-
prior to the measurement of outcome; (5) the study could not be an volved an eating disorder sample.
analogue study; (6) the study needed to be available in English; (7) The overall weighted mean correlation was .22 (k ⫽ 16, n ⫽
the study must have included at least 10 participants; and (8) if the 1306, p ⬍ .001) with a 95% confidence interval of ⫹/⫺.06. This
study did not directly report a correlation between alliance and estimate is very similar to the result obtained by Shirk and Karver
outcome, enough information had to be available in the manuscript (2003) for associations between youth relationship measures and
to calculate the effect size. For studies that included both concur- treatment outcomes (rw ⫽ .21). The result is quite comparable to
rent and prospective associations, only the prospective data were an estimate (rw ⫽ .19) based on a broader sample of child and
included in analyses. adolescent studies that included both concurrent and prospective
 SPECIAL ISSUE: ALLIANCE IN YOUTH THERAPY 21

Table 1
Reviewed Studies, Treatment Type, Alliance Type, Problem Type, Sample Size, and Effect Sizes

Study Treatment type Alliance type Problem type N E.S.

Adler (1998) Non-behavioral Parent Mixed problems 92 24

Auerbach, May, Stevens, & Kiesler (2008) Non-behavioral Adolescent Substance abuse 39 .12
Chui, McLeod, Har, & Wood (2009) Behavioral Child Internalizing 34 .21
Creed (2006) Behavioral Adolescent Internalizing 68 .31
Darchuck (2007) Non-behavioral Adolescent Substance abuse 40 .26
Eltz, Shirk, & Sarlin (1995) Non-behavioral Adolescent Mixed problems 38 .08
Green et al. (2001) Non-behavioral Child & Adolescent Mixed problems 55 .08
Non-behavioral Parent Mixed problems 55 ⫺.08
Hintikka, Laukkanen, Marttunen, & Lehtonen (2006) Non-behavioral Adolescent Mixed problems 45 .07
Hogue, Dauber, Stambaugh, Cecero, & Liddle (2006) Behavioral Adolescent Substance abuse 56 ⫺.02
Karver et al. (2008) Behavioral Adolescent Internalizing 11 .34
Non-behavioral Adolescent Internalizing 12 ⫺.15
Kazdin, Marciano, & Whitley (2005) Behavioral Child Externalizing 75 .39
Behavioral Parent Externalizing 185 .21
Kazdin, Whitley, & Marciano (2006) Behavioral Child Externalizing 77 .32
Behavioral Parent Externalizing 77 .28
Kazdin & Whitley (2006) Behavioral Parent Externalizing 218 .23
McLeod & Weisz (2005) Non-behavioral Parent Internalizing 21 .30
Non-behavioral Child Internalizing 22 .21
Shirk et al. (2008) Behavioral Adolescent Internalizing 50 .27
Zaitsoff, Doyle, Hoste, & Le Grange (2008) Non-behavioral Adolescent Eating disorders 36 .53

Note. E.S. ⫽ effct size.

designs as well as family, group, and milieu therapy in addition to The File Drawer Problem
individual therapy (Shirk & Karver, in press).
Effect sizes ranged from ⫺.08 to .53, yet, the Q statistic did not One of the criticisms of meta-analysis is the “file drawer prob-
indicate statistically significant variability across studies, Q(15) ⫽ lem” (Rosenthal, 1979). It is commonly suspected that studies that
18.29, p ⫽ .25. However, given the very small number of effect yield statistically significant results with larger effect sizes are
sizes, it is clear that the homogeneity statistic lacked adequate more likely to be published than studies yielding smaller and/or
power to detect systematic variability (Rosenthal, 1995). Thus, nonsignificant effects; these remain in the “file drawer.” Thus, it is
exploratory analyses were conducted to determine if variability in possible that the reported weighted mean effect size could be
the effect sizes was moderated by other variables. subject to potential upward bias. To examine this possibility, we
calculated the fail-safe N (FSN). The FSN estimates the number of
unpublished studies with effect sizes of zero that would be needed
Moderators
to reduce the overall effect size to a less meaningful level (Lipsey
Alliance– outcome associations estimated by age of client & Wilson, 2001; Orwin, 1983; Rosenthal, 1991). For this analysis,
showed some variability with weighted mean correlations of .32 we set the criterion effect size level to .10 (small effect). We
(CI ⫽ ⫹/⫺.14) for children, .19 (CI ⫽ ⫹/⫺.10) for adolescents, utilized the formula K0 ⫽ k [ESk/ESc ⫺ 1]. K0 ⫽ the number of
and .21 (CI ⫽ ⫹/⫺.08) for parents. Given the small sample size effect sizes equal to zero needed to reduce the mean effect size to
and the relatively broad confidence intervals, these estimates did ESc; k is the number of effect sizes in the current meta-analysis;
not attain statistical significance ( p ⬎ .05). The average weighted ESk is the weighted mean effect size; ESc is the criterion effect size
mean correlations for behavioral and nonbehavioral treatments level. Using ESc ⫽ .10, it was found that 19 additional studies with
were .25 (CI ⫽ ⫹/⫺.07) and .16 (CI ⫽ ⫹/⫺.10), respectively. null effects would be needed to be found to reduce the present
Given the large confidence interval for nonbehavioral therapies, meta-analysis overall effect size to .10. Given that this is larger
the apparent difference in alliance strength cannot be viewed as than the number of studies in the current meta-analysis, one which
reliable ( p ⬎ .05). Finally, alliance– outcome correlations varied already contains five studies (out of 16) with effect sizes below
across type of treated problems with weighted mean correlation of .10, it seems unlikely that there is a file drawer problem influenc-
.10 (CI ⫽ ⫹/⫺.17) for substance abuse, .25 (CI ⫽ ⫹/⫺.14) for ing the overall effect size found in this meta-analysis.
internalizing problems, .26 (CI ⫽ ⫹/⫺.12) for externalizing prob-
lems, .10 (CI ⫽ ⫹/⫺.12) for mixed problems, and .53 (CI ⫽ Discussion of Findings
⫹/⫺.34) for eating disorders. However, given large confidence
intervals most differences are not reliable ( p ⬎ .05). However, The current meta-analysis of prospective studies of individual
based on these confidence intervals the estimate for the eating youth alliance– outcome associations yielded an effect that is quite
disorder sample differed from both the substance abuse and mixed comparable to results obtained in the adult literature (Martin et al.,
problems samples ( p ⬍ .05). It should be noted that only one study 2000). It is noteworthy that this is the first meta-analysis to use
evaluated alliance– outcome relations with an eating disordered inclusion criteria that are highly similar to those used in adult
sample. meta-analyses. Specifically, the sample was restricted to studies of
22 SHIRK, KARVER, AND BROWN

individual therapy that measured alliance prior to outcome. Al- Research on the alliance in child and adolescent therapy could
though the establishment of temporal ordering is essential for be improved in several ways. First, although the emergence of
causal inference, it is not sufficient. Nevertheless, these results prospective studies in the youth literature represents an important
strengthen the claim that the alliance is an important predictor of advance over early research, no study has controlled for the impact
treatment outcome in child and adolescent therapy. Future studies of early improvement on alliance. Given evidence for early gains
need to account for the potential impact of treatment gains prior to in youth therapy, direction of effect remains unclear even with
alliance measurement, and for other process variables that could prospective designs. Second, virtually all studies have examined
share predictive variance with the alliance. direct relations between alliance and outcome even though one of
A statistical test of effect size heterogeneity failed to produce a the most prominent theoretical models posits an indirect link
significant result. However, because small sample size limited between alliance and outcome. Neither indirect nor mediation
power in this analysis, and because prior youth meta-analyses models have been tested in the youth literature. Finally, the con-
revealed meaningful variability (Shirk & Karver, 2003), a limited tribution of the alliance to outcome tends to be evaluated in
set of moderators was explored. Results showed some variability isolation. Thus, the unique contribution of alliance relative to other
in alliance– outcome relations across age groups, but these findings process predictors such as client involvement in specific tasks or
did not attain statistical significance. It was somewhat surprising to therapists use of specific techniques remains unknown.
see a marginally stronger association between alliance and out-
come for children than for adolescents. Clinical observations have Therapeutic Practices
underscored the importance of the alliance in adolescent treatment
(Castro-Blanco & Karver, 2010), but current results suggest that The alliance has a long history in the child and adolescent
the alliance is a robust predictor of outcomes with preadolescent literature. Recent research progress on alliance– outcome relations
children. It is possible that these apparent age-related trends are a indicates that this long-standing interest is clearly justified. Alli-
function of differences in the types of problems treated in these ance is an important predictor of youth therapy outcomes and may
two age groups. Specifically, studies with substance abusing youth very well be an essential ingredient that makes diverse child and
produced small effects and all of these studies were conducted adolescent therapies work.
with adolescents. } Alliances with both youth and their parents are predictive of
There was a trend for alliance– outcome associations to be treatment outcomes. Consequently, psychotherapists need to at-
stronger in behavioral than nonbehavioral therapies. Although this tend to the development of multiple alliances, not just to the
result was not statistically reliable, it was still surprising. Nonbe- alliance with the youth. A solid alliance with the parent may be
havioral youth therapies have emphasized the importance of the particularly important for treatment continuation.
alliance for beneficial outcomes, whereas behavioral treatments } Parents and youth often have divergent views about treatment
have focused on specific technical procedures. Current results goals. Formation of a therapeutic alliance with both youth and
parent requires the therapist to attend to multiple perspectives
suggest that the alliance is important for outcome in both types of
and to develop a treatment plan that accommodates both youth and
therapy.
parent perspectives.
Finally, there was some variability in alliance– outcome associ-
} The maintenance of a positive alliance over time predicts
ations across types of treated problems, however only the differ-
successful outcomes with youth. Therapists are advised to monitor
ence between the eating disorder sample and both substance abuse
alliance over the course of treatment. Alliance formation is not
and mixed problems samples proved to be reliable. This finding
simply an early treatment task, it is a recurrent task.
should be viewed with caution as only one study evaluated
} Youth are likely to have a limited understanding of therapy.
alliance– outcome relations with eating disordered youth. It is
Initial results suggest that early alliance formation with youth
likely that this estimate will change as additional studies are
requires the therapist to balance active listening to the youth with
completed. It was somewhat surprising to find such a small asso-
providing an explicit framework for understanding therapy pro-
ciation between alliance and outcome with substance abusing
cesses (roles, tasks, relevance). Overemphasizing the latter to the
youth. Clinical theory has emphasized the importance of motiva-
exclusion of the former appears to interfere with alliance forma-
tional interventions to secure a working alliance with this group
tion, at least with adolescents.
(Miller & Rollnick, 2002). It is not clear if this result reflects
unique difficulties with alliance assessment with substance abus-
ing youth, or an accurate estimate of the limited contribution of References
alliance to outcome with substance-abusing teens. References marked with an asterisk indicate studies included in the meta-
analysis. The in-text citations to studies selected for meta-analysis are
not preceded by asterisks.
Limitations of the Research
*Adler, A. G. (1998). The alliance between child therapist and parent:
The field has yet to reach consensus about the crucial features of How it predicts treatment outcome and reflects parent characteristics.
Unpublished doctoral dissertation, Yale University, New Haven, CT.
the therapeutic alliance with youth. We were not surprised, then, to
*Auerbach, S., May, J., Stevens, M., & Kiesler, D. (2008). The interactive
find a large number of alliance measures in a relatively small role of working alliance and counselor-client interpersonal behaviors in
number of studies. One way to advance the field would be to adolescent substance abuse treatment. International Journal of Clinical
conduct a study in which multiple measures of alliance are admin- and Health Psychology, 8, 617– 629.
istered in order to derive core underlying dimensions. Of course, Axline, V. (1947). Play therapy. Boston: Houghton Mifflin.
sensitivity to developmental issues should guide such an endeavor. Bordin, E. S. (1979). The generalizability of the psychoanalytic concept of
 SPECIAL ISSUE: ALLIANCE IN YOUTH THERAPY 23

the working alliance. Psychotherapy: Theory, Research, and Practice, *Karver, M., Shirk, S., Handelsman, J. B., Fields, S., Crisp, H., Gudmund-
16, 252–260. sen, G., & McMakin, D. (2008). Relationship processes in youth psy-
Castro-Blanco, D., & Karver, M. (2010). The elusive alliance: Treatment chotherapy: Measuring alliance, alliance-building behaviors, and client
engagement strategies with high-risk adolescents. Washington, DC: involvement. Journal of Emotional and Behavioral Disorders, 16, 15–
APA Books. 28.
*Chiu, A. W., McLeod, B. D., Har, K., & Wood, J. J. (2009). Child- *Kazdin, A. E., Marciano, P. L., & Whitley, M. K. (2005). The therapeutic
therapist alliance and clinical outcomes in cognitive behavioral therapy alliance in cognitive-behavioral treatment of children referred for oppo-
for child anxiety disorders. Journal of Child Psychology and Psychiatry, sitional, aggressive, and antisocial behavior. Journal of Consulting and
50, 751–758. Clinical Psychology, 73, 726 –730.
*Creed, T. (2007). A mediation model of early predictors of treatment *Kazdin, A. E., Whitley, M., & Marciano, P. L. (2006). Child-therapist and
outcome within cognitive-behavioral therapy for children with anxiety parent-therapist alliance and therapeutic change in the treatment of
disorders: Child involvement, therapist behavior, and alliance. Disser- children referred for oppositional, aggressive, and antisocial behavior.
tation Abstracts International: Section B: The Sciences and Engineer- Journal of Child Psychology and Psychiatry, 47, 436 – 445.
ing, 68(3-B), 1919. *Kazdin, A. E., & Whitley, M. K. (2006). Pretreatment social relations,
*Darchuk, A. J. (2007). The role of the therapeutic alliance and its therapeutic alliance, and improvements in parenting practices in parent
relationship to treatment outcome and client motivation in an adolescent management training. Journal of Consulting and Clinical Psychology,
substance abuse treatment setting. (ProQuest Information & Learning). 74, 346 –355.
Dissertation Abstracts International: Section B: The Sciences and En- Kendall, P., Comer, J., Marker, C., Creed, T., Puliafico, A., Hughes, A., &
gineering, 68 (5-B), 3392. Hudson, J. (2009). In-session exposure tasks and therapeutic alliance
DiGiuseppe, R., Linscott, J., & Jilton, R. (1996). Developing the therapeu- across the treatment of childhood anxiety disorders. Journal of Consult-
tic alliance in child-adolescent psychotherapy. Applied and Preventive ing and Clinical Psychology, 77, 517–525.
Psychology, 5, 85–100. Linscott, J., DiGiuseppe, R., & Jilton, R. (1993). A measure of TA in
*Eltz, M., Shirk, S., & Sarlin, N. (1995). Alliance formation and treatment adolescent psychotherapy. Poster presented at the 101st Annual Con-
outcome among maltreated adolescents. Child Abuse & Neglect, 19, vention of the American Psychological Association, Toronto, Canada.
419 – 431. Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. Thousand
Estrada, A., & Russell, R. (1999). The development of the child psycho- Oaks, CA: Sage.
therapy process scales (CPPS). Psychotherapy Research, 9, 154 –166. Martin, D. J., Garske, J. P., & Davis, M. K. (2000). Relation of the
Faw, I., Hogue, A., Johnson, S., Diamond, G. M., & Liddle, H. A. (2005). therapeutic alliance with outcome and other variables: A meta-
The Adolescent Therapeutic Alliance Scale: Development, initial psy- analytic review. Journal of Consulting and Clinical Psychology,
chometrics, and prediction of outcome in family-based substance abuse 68(3), 438 – 450.
prevention counseling. Psychotherapy Research, 15, 141–154. *McLeod, B., & Weisz, J. (2005). The therapy process observational
Freud, A. (1946). The psychoanalytic treatment of children. New York: coding system-alliance scale: Measure characteristics and prediction of
International Universities Press. outcome in usual clinical practice. Journal of Consulting and Clinical
*Green, J., Kroll, L., Imrie, D., Frances, F., Begum, K., Harrison, L., & Psychology, 73, 323–333.
Anson, R. (2001). Health gain and outcome predictors during inpatient Miller, W. R., & Rollnick, S. (2002). Motivational interviewing: Preparing
and related day treatment in child and adolescent psychiatry. Journal of people for change. New York: Guilford Press.
the American Academy of Child & Adolescent Psychiatry, 40, 352–332. Orwin, R. (1983). A fail-safe N for effect size. Journal of Educational
Hawley, K., & Weisz, J. (2003). Child, parent and therapist (dis)agreement Statistics, 8, 147–159.
on target problems in outpatient therapy: The therapist’s dilemma and its Rogers, C. R. (1957). The necessary and sufficient conditions of therapeu-
implications. Journal of Consulting and Clinical Psychology, 71, 62–70. tic personality change. Journal of Consulting Psychology, 22, 95–103.
Hawley, K. M., & Weisz, J. R. (2005). Youth versus parent working Rosenthal, R. (1979). The “file drawer problem” and tolerance for null
alliance in usual clinical care: Distinctive associations with retention, results. Psychological Bulletin, 86, 638 – 641.
satisfaction and treatment outcome. Journal of Clinical Child and Ad- Rosenthal, R. (1991). Meta-analytic procedures for social research (rev.
olescent Psychology, 34, 117–128. ed.). Newbury Park, CA: Sage.
Hedges, L. V. (1994). Fixed effects models. In H. Coooper & L. Hedges Rosenthal, R. (1995). Writing meta-analytic reviews. Psychological Bul-
(Eds.), The handbook of research synthesis (pp. 285–299). New York: letin, 118, 183–190.
Russell Sage Foundation Foundation. Sandler, J., Kennedy, H., & Tyson, R. (1980). The techniques of child
Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. psychoanalysis: Discussions with Anna Freud. Cambridge, MA: Har-
New York: Academic Press. vard University Press.
*Hintikka, U., Laukkanen, E., Marttunen, M., & Lehtonen, J. (2006). Good Shirk, S. (1988). Cognitive development and child psychotherapy. New
working alliance and psychotherapy are associated with positive changes York: Plenum Press.
in cognitive performance among adolescent psychiatric inpatients. Bul- *Shirk, S., Gudmundsen, G., Kaplinski, H., & McMakin, D. (2008).
letin of the Menninger Clinic, 70, 316 –335. Alliance and outcome in cognitive-behavioral therapy for adolescent
*Hogue, A., Dauber, S., Stambaugh, L., Cecero, C., & Liddle, H. (2006). depression. Journal of Clinical Child and Adolescent Psychology, 37,
Early therapeutic alliance and treatment outcome in individual and 631– 639.
family therapy for adolescent behavior problems. Journal of Consulting Shirk, S., & Karver, M. (2003). Prediction of treatment outcome from
and Clinical Psychology, 74, 121–129. relationship variables in child and adolescent therapy: A meta-
Horvath, A. O., & Greenberg, L. S. (1989). Development and validation of analytic review. Journal of Consulting and Clinical Psychology, 71,
the Working Alliance Inventory. Journal of Counseling Psychology, 36, 462– 471.
223–233. Shirk, S., & Karver, M. (2006). Process issues in cognitive-behavioral
Karver, M., Handelsman, J., Fields, S., & Bickman, L. (2006). Meta- therapy for youth. In P. C. Kendall (Ed.), Child and adolescent therapy:
analysis of relationship variables in youth and family therapy: Evidence Cognitive-behavioral procedures (pp. 465– 491). New York: Guilford
for different relationship variables in the child and adolescent treatment Press.
literature. Clinical Psychology Review, 26, 50 – 65. Shirk, S., & Karver, M. (in press). Alliance in youth psychotherapy. In J. C.
24 SHIRK, KARVER, AND BROWN

Norcross (Ed.), Psychotherapy relationships that work (2nd ed.). New Wright, L., Everett, F., & Roisman, L. (1986). Experiential psychotherapy
York: Oxford University Press. with children. Baltimore: Johns Hopkins University Press.
Shirk, S., & Russell, R. L. (1996). Change processes in child psychother- *Zaitsoff, S., Doyle, A., Hoste, R., & le Grange, D. (2008). How do
apy: Revitalizing treatment and research. New York: Guilford Press. adolescents with bulimia nervosa rate the acceptability and therapeutic
Shirk, S., & Russell, R. L. (1998). Process issues in child psychotherapy. relationship in family-based treatment? International Journal of Eating
In A. Bellack & M. Hersen (Eds.), Comprehensive clinical psychology. Disorders, 41, 390 –398.
London: Pergamon Press.
Shirk, S., & Saiz, C. (1992). Clinical, empirical, and developmental per- Received August 9, 2010
spectives on the therapeutic relationship in child psychotherapy. Devel- Revision received October 19, 2010
opment and Psychopathology, 4, 713–728. Accepted October 19, 2010 䡲