8/16/2019 Alliance Brief
1/22
Levels and Pat terns of the Therapeutic
Alliance
in
Brief Psychotherapy
CHRISTOPHER L. STEVENS, Ph.D.*
J. CHRISTOPHER MURAN, Ph.D.*
JEREMY D. SAFRAN, Ph.D.
BERNARD S. GORMAN, Ph.D.f .. •
ARNOLD WINSTON, M.D.*
We examined the relevance of the level and pattern of the therapeutic
alliance in 44 cases of three different manualized 30-session treatments
using patient
r tings
of the Working Alliance Inventory after
e ch
session. It
was hypothesized that bo th high-alliance level and either linear incre se in
alliance rating or a series of hrief rupture-and-repair episodes would be
found
in
successful treatments.
We
also hypothesized that
a
more global
high-low-high pattern predicted
in tbe
literature would
not be
present.
Consistent with the literature higher lli nce levels were found to be related
to improved
outcome. As
predicted,
we
did not find
glohal, high-low-high
pattern. Local rupture-and-repair patterns were found in 30 of the cases;
linear trends were found in 66 of the cases. There was no relationship
between outcome
and
either pattern.
We
found
no
differences among
the
treatments.
LEVEL PATTERN OF THE THER APEUT IC ALLL^NCE IN BRIEF
PSYCHOTHERAPY
Although findings that a strong, positive therapeutic alliance is related
to positive treatment outcome has been quite consistent, (Horvath and
Symonds, 1991; Martin, Garske, & Davis, 2000) the exact nature of this
relationship has remained less clear. As a result, clinical theorists and
researchers have increasingly turned their attention to determining how
the alliance functions to foster change. For example, some (e.g., Raue &
*Beth Israel Medical Center
and
Alben Einstein College
of
Med icine, New School
for
Social
8/16/2019 Alliance Brief
2/22
AMERICAN JOURNAL O F PSYCH OTHER APY
Goldfrie d, 1994; Beck, 1995) believe that the alliance serves prim arily as a
static foundation for the application of curative techniques, while others
(e.g., Bo rdin, 1979, 1994; Safran M ura n, 2000) believe that the alliance
is an active agent of the change process in its own right. Many of these
authors have theorized that the way in which the alliance develops over
time can give important information about how this takes place. More
specifically, they hypothesized that working through conflicts or strains in
the alliance leads to positive change and that the variability in the alliance
ratings, indicating these periods of strain, could be a pre dictor of impro ved
outcome.
Much of this research can be characterized as a search for global shifts
in the alliance over the course of treatme nt. Researchers, influenced in pa rt
by Mann (1973), suggested that alliance developm ent goes throu gh a series
of changes as patients react to the shifting techniques of the therapist, (e.g.
Ge lso Ca rter 1985, 1994; H orv ath Lub orsky, 1993, Kivlighan
Shaughnessy, 1995, 200 0). Specifically, resea rchers posit that th e alliance
will proceed through three distinct phases. According to this model, the
alliance is characterized by an initial period of high hope fostered by
supportive alliance-building activity by the therapist. During the next
phase of treatm ent, the patient realizes that all expectations will not be m et
and increasingly challenges therapist activity, which leads to growing
ambivalence, resistance, and greater strain on the alliance. Finally, in
successful treatment, patients gain a more realistic understanding of the
therapist's role and accept change and the limitations of therapy, which is
brought about (at least in part) by the therapist's focus on working through
and termination. This process may be perceived as following a high-low-
high pattern of alliance development.
A number of studies tried to demonstrate this global high-low-high
pa ttern with mixed results (Bernard, Schw artz, Ocaltis Stiner, 1980;
G olden Rob bins, 1990; H orvath M arx, 1991; Joyce an d P iper 1990;
Kivlighan Shaughnessy, 2000; Pa tton , Kivlighan M ulton , 1997;
Schwartz and B ernard,
1981;
and Tracey, 1989). Th e most com pelling case
for a high-low-high pattern comes from KivHghan and Shaughnessy
(2000). They noted that as a result of the statistical procedures used, both
their 1995 study and the study by Patton, Kivlighan, and Multon (1997)
were unable to detect possible subgroups of alliance development. As a
8/16/2019 Alliance Brief
3/22
Alliance Level and Pattem
come. They used novice counselors and recruited subjects for a 4-session,
short-term dynamic therapy. Analysis of their initial sample of 38 clients
showed three distinct patterns of alliance development, which they iden-
tified as stable alliance, linear alliance growth, and quadratic alliance
growth. A subsequent replication sample of 41 clients identified the same
three p attern s. In add ition, this analysis of the results of the second sample
showed that clients reporting a quadratic pattern of alliance development
had greater improvement on outcome measures compared to other pat-
terns of development.
Although the results offer the strongest support to date of a rupture-
and-repair process, there are several significant limitations to this study.
The stated goal of the study was to examine patterns of global alliance
development, but such a brief treatment is a significant shortcoming. First,
it seems unlikely that any of the stages that Mann described had time to
develop. Similarly, Horvath and Luborsky's (1993) model also suggested
that the alliance-building phase of treatment would take place over the
first five sessions, peaking at session three. With only four sessions of
treatment, it is impossible to determine if the changes observed were
related to those that might be seen in longer treatment. Second, with only
four sessions, there are otily 16 potential patterns that could develop,
several of which could be interpreted as high-low-high, stable, or linear
improvement. Since Kivlighan and Shaugnessy found three different pat-
terns, it might be more important to ask which patterns did not emerge.
And finally, it is impossible to distinguish between discrete or transient
changes in alliance and more global dynamics.
Although, taken as a group, these studies seem to provide support for
the notion of a global high-low-high process, all suffered from significant
limitations. Sample sizes were small or not representative of clinical
pop ulation s (Tracy, 1989; G old en Robb ins, 1990). The training and
exp erien ce of the therapis ts involved was limited (Kivlighan Shaugh-
nessy, 1995, 2000 ; Pa tton et al., 1997). Tre atm ent length was inconsistent
(Bernard et al. 1980; Bernard and Schwartz, 1981) or extremely brief
(Kivlighan Shaughnessy, 2000). Add itionally, the statistical m ethod s
used were not appropriate for detecting patterns that might exist among
subg roup s of patien ts (Tracy, 1989; G old en Robb ins, 1990; Kivlighan
Shaughnessy, 1995; Patton et al., 1997). These limitations raise the ques-
8/16/2019 Alliance Brief
4/22
A M E R IC A N J O U R N A L O F P S Y C H O T H E R A P Y
A review of the literature to date suggests that there is moderate
support for the notion that variability increases in the midphase of
treatment (Bernard, Schwartz, Ocaltis and Stiner, 1980; Schwartz and
Bernard, 1981). Similarly, there is some initial support for the notion that
undergoing a rupture and repair process is related to improved outcome
(Form an M arm ar, 1985; Lansford, 1986; G olde n Robb ins, 1990;
H orva th M arx, 1991; Kivlighan Shaughnessy, 2000 ; Pa tton , Kivlighan
M ulton, 1997). How ever, Ge lso and Carter, as well as H orva th and
Luborsky, point out that the strains in the alliance experienced during the
middle phase of treatment often are the result of more focal strains on the
relationship, which are brought about by momentary lapses or empathic
failures on the part of the therapist, disagreements on tasks or goals, or
strains on the bo nd. Both empirical (Forman M armar, 1965; Rho des,
H ill, Th om pson Elliott, 1994; Safran, Cro cker, M cM ain M urray,
1990; M uran , 2002; Safran M uran 1996; Safran, M uran Samstag,
1994; Safran, M uran , Samstag Stevens, 2002 in press) and theoretical
(Bord in, 1994; Safran M ura n, 2000) work suggest that when strains in
the alliance are not addressed, outcome is negatively impacted and drop
out may occur.
If this is true, then the middle phase of treatment might be better
characterized by a series of brief ruptures and repairs in which the
alliance is strained. In successful treatment, the strain is addressed and
the alhance returns to previously high levels, not during the final stage
of treatment, but within a few sessions. Rather than thinking of alliance
development in terms of curves, with global decrease in patient satis-
faction and general weakening of the alliance in the middle phase, it
could be envisioned as a series of dips and spikes representing, not
global shifts in satisfaction, but a kind of session-by-session negotia-
tion proces s (Safran M ura n, 200 0).
The present study attempts to address many of the methodological
problems that limited the conclusions drawn from previous studies and to
test the idea that the rupture-and-repair process is better described as a
local, rather than global, phenomenon. Consistent with the Bordin's
(1994) assertion that different patients work differently in the alliance, two
patterns of alliance development were predicted. The first form is a linear
increase in Working Alliance Inventory (WAI) scores across the course of
8/16/2019 Alliance Brief
5/22
Alliance Level and Pattern
to five sessions. A global high-low-high pattern was not expected to
appear. Additionally, it was predicted that treatments demonstrating a
linear increase or rupture-and-repair episodes would evidence better
outcome than those that did not.
METHODS
PARTICIPANTS ., . . -
her pists
This study was conducted as part of a psychotherapy research project
at a major medical center in New York City (see Muran, 2002, for an
overview). The therapists were drawn from the department of psychiatry
and included attending psychiatrists, clinical psychologists, social workers,
psychiatry residents, psychology interns and externs. To assure consistency
of treatment, all of the therapists, regardless of their background and
experience, received similar training and evaluation in the therapeutic
model they practice. To assure treatment adherence, trained observers
rated videotapes of a randomly selected session in each treatment third
(beginning, middle and end), using an instrument developed by the
project, which has demonstrated adequate psychometric properties (Pat-
ton, M uran, Safran, W ach tel, W insto n, 1996). Licensed clinicians
participated in one-hour weekly supervision until they met acceptable
standards for treatment adherence (1-2 cases), while unlicensed therapists
continued to receive individual supervision for all of their cases. All
therapists participated in a weekly 90 -m in ut e g roup supervision.
Patients were randomly assigned to one of three manualized, 3 0 -
session, on ce- pe r-w eek treatment conditions designed to treat person ality-
disordered patients. All the therapy sessions were videotaped. The three
different conditions were Brief Relational Therapy ([BRT] Safran, 2002;
Safran M uran , 2000; M uran Safran, 2002 ), Co gnitive-Behavioral
Therapy ([CBT ] Tu rner M uran, 1992) and Brief Ad aptive Psychother-
apy ([B AP]; Pollack, Flegen heirmer Kaufman, 1988). Brief Relational
Therapy is an integrative model that combines the principles of relational
psychoanalysis and humanistic psychotherapy. It focuses on detection and
resolution of alliance ruptures and has an explicit emphasis on process
rather than content. The CBT condition is a fairly traditional, schema-
8/16/2019 Alliance Brief
6/22
AMERICAN JOURNAL OF PSYCHO THERA PY
to the environment in a more adaptive fashion. Of the 128 cases reviewed
for inclusion in this the project, 44 fit the specific inclusion criteria
(presented below). Of the 44 patients, eighteen were seen in CBT, 14 in
BRT and 12 in BAP.
Patients
Patients were recruited using advertisements in local newspapers,
inviting adults suffering from long-standing depression, anxiety, or
interpersonal problems to participate in a program to investigate
short-term psychotherapy. The principal inclusion criterion was that
they met criteria for diagnosis of Personality Disorder (PD) cluster C or
NOS on axis II. Exclusion Criteria was evidence of schizophrenia or
other psychoses, organic brain syndrome or mental retardation, mania
or bipolar disorder, severe obsessive compulsive disorder or serious
eating disorder, serious dissociative disorders, a current substance use
disorder, active para-suicidal or suicidal behavior, a history of violent
behavior or severe impulse control problems, psychotropic medication
use within the past six months.
Patient
demographics.
Th ere were an equal num ber of male and female
patients. They ranged in age from 25 to 63 years (M = 40, SD =10). All
w ere edu cated at or above the high school level, and 7 5 had a college
degree or at least some gradu ate level education. M ore than 7 5 were
employed. Eighty percent were Caucasian.
Patient diagnostic characteristics.
All of the patients in this study m e
the criteria for an Axis II diagnosis, and all but two met criteria for an Axis
I diagnosis in the
DSM IV
(American Psychiatric Association, 1994). All
patients with an Axis I diagnosis were in the mood or anxiety disorder
spectrum. All 44 patients met criteria for an Axis II diagnosis, and 14 also
met criteria for a secondary Axis II diagnosis based on the SCID-II. More
than half (23) had a personality disorder NOS, with the remainder
diagnosed as falling into Cluster C (14 avoidant, 7 obsessive compulsive,
aggressive).
Coders
We had four coders classify alliance development curves and identify
8/16/2019 Alliance Brief
7/22
Alliance Level and Pattern
MEASURES
lliance
Alliance was assessed using the short form of the WAI (Horvath
G reenb erg, 1989, Tracey Kokotovic, 1989).The W AI was adm inistered
at the end of each session as part of a postsession questionnaire (PSQ)
designed to assess the therapeutic relationship (Muran et. al. 1991).
The WAI (Horvath, 1989) was created to assess the working alliance
independent of a therapist's theoretical orientation while simultaneously
providing a clear description of what constitutes a working alliance and
how the alliance functions to promote positive change. Items for the WAI
were generated in a unique multi-step process to insure content validity.
Items were created initially to reflect the three different dimensions (tasks,
goals and bond) of Bordin's model (Bordin 1974, 1979, 1980) and were
rated by psychologists from different theoretical backgrounds. Experts on
the working alliance reviewed each item, rating it for relevance. The
remaining items were reviewed a second time from randomly selected
practicing psychologists and separated into clusters corresponding to each
of the three dimensions (tasks, goals and bo nd ). Th e to p- rat ed 12 items in
each group were chosen to form the WAI. There has been good empirical
support both for the overall scale and the subscales.
Tracey and Kokotovic (1989) used a hierarchical bilevel model to
represent the factor structure of the WAI, and they concluded that the
WAI assesses both the individual aspects of the alliance represented by the
subscales and the overall alliance dimension. Based on their findings, they
selected the four items that best defined each of the unique aspects of the
alliance (i.e. the three subscales) and constructed a 12-item short form of
the WAI. It is this shortened version that is used as a part of the PSQ and
provides the alliance ratings used in this study.
utcome
me sures
Outcome for this study was measured using two factors established by
calculating the standardized residual gains for each of six outcome mea-
sures,
the Global Assessment Scale ([GAS] Endicott, Spitzer, Fleiss,
Cohen 1976), Global Symptom Index (GSI) of the Symptom Checklist
90-Revised ([SCL—90—R] Derogatis, 1983), Inventory of Interpersonal
Problems
([IIP] ,
H oro w itz, Ro senberg, Baer, U reno Villasefior, 1988),
8/16/2019 Alliance Brief
8/22
AMERICAN
J O U R N A L
OF
P S Y C H O T H E R A P Y
analysis with Varimax rotation.
Two
factors were ex tracted with eigenval-
ues exceeding
1.00 and
with
5 9 . 1 1 % of the
variance accounted.
Two
outcome composites were calculated
by
averaging
the
standardized scores
of
the
measures that loaded
> .45 on the
respective factors
and by
applying the yielded factor scores as weights. Factor one, which is consid-
ered
a
m easure
of
symptom reduction,
is a
composite that included
GAS
(factor loading
=
-.78, factor score
=
- .36) ,
PTC
(.83, .41),
TRC
(.81,
.40)
and
the GSI {.45, .12).
Factor
two, a
measure
of
patients interpersonal
functioning,
is a
composite that included
the IIP
(.82,), WIPSI
.89, .61)
and
GSI .50, .23).
STATISTICAL METHODOLOGY
ata selection
Since patterns
of
alliance development were
the
focus
of
this study,
gaps
in PSQ
reporting could have been problematic.
As a
result,
we
included only cases
in
which
the
patient
had
completed treatment,
and
returned
a
minimum
of
6 6 %
of the
PSQ s, with
no
gaps
of no
more than
three consecutive sessions.
Of the 128
original cases,
44 met all of
these
criteria
and
were included
in the
analysis.
The
largest
gap in
consecutive
sessions
was
three , w ith
an
average
of
one missed session.
To
allow
for the
consistent analysis
of
patterns across
the
length
of
therapy,
all
gaps
in PSQ
scores were filled
in
with
the
average
of
scores from
the
sessions be fore
and
after
the
skipped session
s). For
example,
if a PSQ
from session
14
were
missing,
the
average
WAI
values
for
sessions
13 and 15
were substituted
for session
14.
As
it
seemed possible that large gaps
in the PSQ
reporting might
be
indicative
of
ruptures,
and
because cases with large numbers
of
missing
PSQ s could
not be
included
in the
analysis,
it is
possible that
the
rem aining
cases might
be
representative
of
only good cases .
To
help control
for
this, the
data
was
analyzed
in
several ways. First,
we
conducted
a
t-test
to
compare the percentage of PSQs returned for clients who completed
treatment with those
who
dropped
out of
treatment.
The
test results were
not statistically significant
/ =
.423,
p
.673). Next,
all
cases
of
patients
who completed treatment were examined
for a
correlation between
the
percentage
of
PSQs returned
and
outcome. Outcom e
for
symptom reduc-
tion and interpersonal functioning both at termination and follow-up was
8/16/2019 Alliance Brief
9/22
Alliance Level and Pattern
Table 1. OVERA LL
W AI
Mean
Standard
Deviation
Ilanjje
Ont-
5.24
0.75
3/>-6.7
WAI MEAN
Phase
Two
5.58
0.72
3.7-7.0
AND STANDARD
Three
5.95
0.69
4 . ^ 7 . 0
Three
5.39
0.86
3.0-7.0
DEVIATION SCORES
Session
Fifteen
5.55
0,85
3.5-7.0
Thirty
6.0
0.75
4.3-7.0
Overall
Treatment
5.59
0.66
4.1-7.0
relationship was found for either symptom reduction or interpersonal
functioning at termination or follow-up. Although the precise meaning of
missing PSQs cannot be determined, the percentage of PSQs returned
does not appear to be predictive either of length of treatment or of
outcome.
Data nalysts
Before analyzing the impact of different patterns of alliance develop-
ment on outcome, we needed to determine first if alliance
level
for this
sample correlated with outcome. The treatment was first divided into
phases, as suggested by the literature (eg., Mann 1973; Gelso and Carter,
1994;
and Horvath and Luborsky, 1993). Using bivariate Pearson corre-
lations with a Spearman two-tailed test of significance, we compared
• aUiance levels at the third , fifteenth, thi rtieth session;
• average alliance levels for p hase on e (sessions on e to five),
• pha se two (sessions six to twen ty-five), and ph ase thre e (twenty six
through thirty); and
• overall W A I mean -.
^
-
with outcomes for symptom reduction (factor one) and interpersonal
functioning (factor two) at term ination (see Table 1 for des crip tion of
statistics). As can be seen in Table 2, the mean WAI ratings for both
factors were correlated significantly with outcome. With the exception of
the ratings in phase one for interpersonal functioning, all of the mean
ratings for the phases were correlated significantly to outcome. Ratings
taken from individual sessions showed a weaker relationship with out-
come, with only sessions 15 and 30 being significant. Interestingly, corre-
lations were highest for phase two and session fifteen, indicating the closest
8/16/2019 Alliance Brief
10/22
AMERICAN JOURNAL OF PSYCHOTHERAPY
Table
2 .
WORKING
ALLIANCE
O ne
BY OUTCOME
Phase
Two Three
CORRELATIONS
Session
Three Fifteen Thirty
Treatment
Mean
Factor Pearson
One Correlation
Sig. (2-tailed)
Factor Pearson
Two Correlation
Sig. {2-tailed)
- . 4 4 4 - . 5 1 6 - . 3 6 4 - . 2 7 7 - . 4 7 9 - . 3 2 7 - . 5 2 4
.006 .000 .024 .092 .002 .044 .000
- . 1 6 7 - . 3 2 9 - . 3 2 5 - . 0 6 3 - . 2 8 1 - . 2 5 5 - . 3 2 5
.280 .030 .032 .682 .064 .094 .032
range, while all of the correlations for factor two, interpersonal function-
ing, were somewhat lower. Low scores for either factor indicate improve-
ment, so that a negative correlation indicates that higher WAI scores are
related to decreased patient symptoms or decreased interpersonal prob-
lems.
Identification of urves
To test the hypothesis that both a linear trend and a series of ruptures
and repairs are predictive of improved outcome, we examined each of the
44 cases to determine if either of these two definitions described the
pattern of alliance development, or if they were better described as
high-low-high or other patterns. Testing was done in several ways. First,
we used the mean WAI scores for each of the 44 cases to construct a
Chance-corrected Coefficient of Proportionality Matrix. This correlation
matrix allows cases to be examined for shifts across a series of cases,
regardless of their abso lute value. This m eans that the sha pe of the alliance
development may be examined independently of the level of the alliance
scores. W e used W ard s m ethod to see if the cases wou ld cluster by pa ttern
of developm ent. W ard s m etho d is a fairly conservative m eth od , which
constructs fairly compact (minimum variance) clusters.
Large breaks in similarity occurred to produce two dusters, each
consisting of four highly similar cases. Examination of the two clusters
indicated that both were forms of linear increase, with the first (r
.46 to
.87) cluster characterized by a steady, but gradual, increase across the
length of treatment with little variation, and we called this pattem linear-
8/16/2019 Alliance Brief
11/22
Alliance Level and Pattern
stable scores, and we labeled this late-linear. No other patterns of alliance
development were identified using the cluster analytic technique.
dentific tion
of Alliance Curves
W e then employed a second m ethod to determine if linear or qu adratic
patterns were present. Both linear and quadratic patterns can be detected
using ANOVAs to describe regression curves. ANOVAs generated for
each model were examined to determine if adding a quadratic term better
described the shape of the alliance over time compared to a simple linear
relationship.
Since many potential curves could be associated with a positive finding
for a quadratic relationship, the coders were asked to examine each of the
curves and to determine if the curves were best described by a high-low-
high, linear, or stable pattern. Four coders examined each of the curves
and classified each as being high-low-high, linear increase, or other
{Kappa = .90). To determine if a case could be categorized as having
undergone a local period of rupture and repair as measured by the WAI,
the same four coders were given graphs of each of the 44 cases and asked
to identify rupture and repair cycles.
A total of 29 cases (66%) were identified as having a linear trend by
regression analysis with ANOVA (significant at the .05 level or better),
indicating that WAI scores improved over the course of treatment. After
examining each of the regression curves, coders identified 13 cases (30%)
as being best categorized as having a linear trend. All of the cases identified
by the coders, and all of the cases in each of the two clusters, were
identified as having a linear trend in the ANOVA analysis.
dentific tion
of Rupture and Repair Events
Rupture-and-repair patterns were identified differently from more
global trends (e.g., linear, high-low-high, etc.). Cases that demonstrated
micro rupture-and-repair events are not easily detected by statistical
procedures. Although these patterns have a distinctive shape, consisting of
a dro p in alliance scores with a subsequ ent return , the length, severity, and
timing of these episodes may be different in each case. It was, therefore,
necessary to use coders to identify these rupture-and-repair episodes.
If we define ruptu res in the therapeutic alliance as disagreements abou t
tasks or goals or prob lems with the relational bond (Safran et al., 2002, and
8/16/2019 Alliance Brief
12/22
AMERICAN JOURNAL OF PSYCHOTHERAPY
F i g u r e 1
LINE GRAPH OF WORKING ALLIANCE RATINGS (MEAN) BY SESSION.
D 2 3 4 S 8 7 3 9 t O I 1 1 2 1 3 1 4 1 5 1 8 1 7 1 8 1 8 2 2 1 2 2 2 3 2 4 2 5 2 6 2 7 2 8 2 9 3 D 3 1
S E S S IO N N U M B E R
and-repair event as a downward shift in WAI ratings from stable (consistent
or dropping no more than 5 point, or an increase in the previous session),
with a drop of one WAI point in one session (moderate) or more than one
point (serious) in one or more consecutive sessions (starting with an initial
drop of or more points), with a subsequent return to within .25 point of the
predrop level or higher within 3 to 5 sessions.
The same four coders who classified regression curves also identified
rupture events. They were able to reliably identify cases as being rupture
free, having only m ode rate rup ture s, or serious rup ture s (K appa ^ .91). A
total of 22 cases (50 ) were cod ed as having rup ture s. Th irteen (3 0 ) of
these were classified as having serious events. See Figure
for an example
of a case with a single rupture-repair episode, which starts in session 16
8/16/2019 Alliance Brief
13/22
Alliance Level and Pattern
for which a linear trend was detected by the initial ANOVAs were
categorized as significant lin ear . Tho se for which a linear trend was
identified by the coders were categorized as co de r-r ate d linear . Because
coders agreed about only two cases as being best described by a global
high-low-high pattern, this category was not included in the final analysis.
Cases identified as having local ru ptu re patte rns were categorized as any
ruptu res . Tho se with only m odera te ruptures were categorized as only
mo derate ruptu res and those with patterns of only serious ruptures were
classified as only serious ru ptures .
We ran ANOVAs, using the General Linear Model of SPSS (9.0), for
each category to test the degree to which they related to outcome for factor
one or factor two. Treatments categorized as linear by ANOVA and coders
were not found to be related significantly to outcome on factor one.
However, cluster 1 (the stable linear group) showed a relationship to
improved outcome on factor one (symptom reduction) with a modest
effect size
{F
(2, 44) ^ 4.497,
p
.045,
r
.31). Similarly, there was no
significant relationship between linear development patterns as defined by
ANOVA, coder or cluster analysis and factor two (interpersonal function-
ing). There also was no relationship between the presence of rupture-and-
repair episodes and outcome on factor two.
Since rupture-and-repair cycles, as operationalized here, were not
significantly related to outcome, a secondary analysis was run to determine
if variability in the alliance was predictive of outcome. To do this, the
standard deviation for each patient's WAI scores was correlated with
outc om e for each factor. A m oderately strong relationship was found
between the standard deviation of the WAI scores and factor one
ir
.345,
p —
.042), indicating that a patie nt's im prove m ent on factor o ne
(symptom reduction) is negatively correlated with variability on alliance.
Factor two (interpersonal functioning) was not found to be significantly
correlated with variability in WAI scores
[r
.028). Additionally, to
explore the different findings between this study and Kivlighan and
Shaugnessy (2000), an additional analysis was conducted. The Kivlighan
and Shaughnessy (2000) study found that treatments with brief high-low-
high patterns were positively correlated with outcome. This study found
no such relationship. We conjectured that this discrepancy might be
8/16/2019 Alliance Brief
14/22
AMERICAN JOURNA L OF PSYCH OTHERA PY
and Shaughnessy study. To determine if this difference in treatment length
had an effect on the relationship between pattern development and
outcome, the WAI scores for session 3 were subtracted from those of
session 30. Eighteen of the 22 cases containing rupture-and-repair events
had higher scores at session thirty than at session three, indicating that a
majority of cases (81 ) with ru pt ur e- an d- re pa ir episodes ended w ith
alliance scores as high or higher than hose with which they began.
Treattnent Conditions
In addition to looking for an overall relationship among alliance level,
pattern of alliance development, and outcome, the data was also analyzed
for relationships among the three treatment conditions and patterns of
alliance development. No significant relationship was found between
alliance levels and treatment group. A series of Chi-Square tests were run
to test for relationships between type of alliance pattern developed with
treatment group. The percentage of cases with ruptures was higher in the
BRT gro up (a total of 10 rup tures in a total of 6 6 of the cases) than in
either CB T or BAP condition (with nine rup ture s in 5 3 of the cases and
3 ruptures in 3 3 of the cases respectively). Th e relationship app roach ed
significance
x^ -
4.725,
p =
.094).
There also was no significant relationship found between linear in-
crease and treatment condition
{)^ —
.625,
p =
.732). There was a
moderately significant relationship
{r
(2, 44) = .38,
p =
.042) between
treatment condition and variability. The BRT condition was found to have
significantly greater variability (M = .66) than BAP (M
=
.51) and CBT
(M = .49). One additional finding is that the linear stable cluster of four
cases found to be related to improved outco m e was m ade u p of three C BT
and one BAP case.
DISCUSSION
The hypothesis that alliance development would be characterized by a
linear increasing pattern and treatments demonstrating localized (three to
five session) rup tu re -an d- re pa ir episodes was sup por ted. As predicted ,
there was no evidence of a global high-low-high pattern of alliance
development and decay. Rather, alliance development was characterized in
8/16/2019 Alliance Brief
15/22
Alliance Level and Pattem
seen in BRT than in BAP or CBT, indicating that rupture-and-repair
events may occur differently in different types of treatment.
The hypothesis that treatments demonstrating a linear trend or rup-
ture-and-repair episodes would be related to improved outcome, received
limited support. No relationship between the presence of rupture-and-
repair episodes and outcome was found. There also was no relationship
found between a statistically significant linear trend or coder-rated linear
trend and outcome, but one cluster of four cases classified as stable-linear
was found to have a significant relationship to improved ou tcom e on factor
one (symptomatology). Finally, variance in WAI ratings was found to be
negatively correlated with outcome on factor one (symptomatology).
There are a number of possible explanations for the failure to find any
significant relationship between the presence of rupture-and-repair epi-
sodes and outcome. First, the way in which ruptures in the alliance were
operationalized in this study may not accurately reflect actual rupture
events. Working from a definition that ruptures in the aUiance are char-
acterized by disagreements about tasks or goals, or problems with the
relational bond, we hypothesized the WAI scores, which dropped and
then returned to pr e -d ro p levels, wo uld reflect this process. Although the
classifications of moderate and serious rupture events were made to
discriminate levels of disagreement, we did not consider the number or
severity of rupture events in this analysis. Currently, there has been no
theoretical prediction about the pattern alliance ruptures make. Safran and
M uran (2000) state that the process of resolving rup ture s m ay, in fact, lead
to further rup ture s. In describing the BRT m odel, Safran w rote, In this
approach, the treatment process is conceptualized as ongoing cycles of
therapeutic enactment, disembedding and understanding, enactment and
disem bed ding (Safran, 2002, p. 171).
Nevertheless, there is no clear notion of how many (or few) rupture
events lead to imp rovem ent. It is possible to imagine cases in which patient
and therapist encounter, and work through, a significant rupture around a
core maladaptive interpersonal issue and then move on to other topics.
Pe rha ps b eing able to survive this one conflict, o r feel understood, just this
once, is enough to lead to significant improvement. It is just as easy to
imagine, however, that a patient may need to work through a particular
conflict several times with a therapist before being able to experience it in
8/16/2019 Alliance Brief
16/22
A M E RICA N J O U R N A L O F P S Y C H O T H E R A P Y
timing, and number of these events, based on a more detailed model,
might yield a better understanding of how the pattern of ruptures may be
related to outcome.
The additional findings, variance in WAI ratings was negatively corre-
lated with outcome, and only the most stable cluster of cases displaying a
linear pattern was correlated with improved outcome, make interpretation
more difficult. Regardless of the specific pattern that alliance development
had been expected to take, be it the global high-low-high pattern or the
more local, rupture-and-repair events predicted here, there is broad
consensus among theorists that fluctuations in the therapeutic relationship
are both unavoidable and form an important component of the change
process. These theorists (e.g. Bordin, 1994; Forman & Marmar, 1965;
H orvath and Luborsky, 1993; M uran 2002; Rhod es, Hill, Th om pson &
Elliott, 1994; Safran, Crocker, McMain & Murray, 1990; Safran & Muran
1994, 1998, 2000; Tracey, 1989) suggest that it is through these disagree-
ments, misunderstandings, and conflicts, that patients learn critical lessons
about how to balance the needs of the self and the needs of others. Tracey
(1989),
and Horvath and Luborsky (1993) suggest that when therapists do
t
notice periods of tension or disagreement, they should begin to
question if the treatment needs to be changed, noting that it is important
to be aware that during periods when therapy goes too sm oothly might
represent a problem in the therapeutic work.
The belief that having the experience of working through conflicts with
an empathic and responsive listener is an essential part of therapeutic
change, reflects years of clinical theory and experience backed by compel-
ling anecdotal evidence. This has lead researchers to attempt to discover
the way therapeutic technique interacts with the alliance and shapes its
development. Finding patterns of alliance development consistent with the
belief that a degree of conflict is an essential element of change has proven
however, to be a difficult task. Studies that have found these patterns (e.g.
Berna rd et al., 1980; Berna rd & Schw artz, 1981; Pa tton et al., 1997;
Kivlighan & Shaughnessy, 2000) have been hampered by methodological
prob lem s. Conversely, the finding that a strong, stable alliance is related to
improved outcome (Joyce & Piper, 1990; Kivlighan & Shaughnessy, 1995;
8/16/2019 Alliance Brief
17/22
Alliance Level and Pattern
little between-session variation is more conducive to improvement than a
more dramatically fluctuating alliance.
There are several possible explanations for this contradiction between
the theory and the findings. First, the shape of the alliance may mean
different things for different pa tien ts. As Bordin 1994) suggests, for som e
patien ts, good outc om e will be associated with a stable alliance. Fo r oth ers,
as Tracey 1989) suggests, an overly stable alliance repres ents collusion
between patient and therapist to avoid difficult topics or over compliance
by the patient. Similarly, the presence of rupture-and-repair events may
mean different things in different treatments. For some patients, these
events may represent the uncovering and working through of maladaptive
interpersonal cycles, for others it may be more indicative of an inability to
form or maintain a consistent relationship or engage in therapeutic work.
If this is true, than rather than looking for globally generalizable patterns,
investigators need to focus on factors that might cause particular dyads to
work in the alliance in a particular way.
Just as alliance patterns vary depending on the particular treatment
dyad, there is some evidence in this study that the shape the alliance takes
relates to treatment condition. There were marginally more cases with
rupture events in the BRT than in BAP and CBT conditions. There was
greater variability in BRT than CBT or BAP cases, and three of the four
members of the stable, high-alliance cluster, which was found to be
significantly related to im proved ou tcom e on factor on e symptom redu c-
tion), were seen in the CBT condition. While these findings are certainly
suggestive, the effect sizes were small and the relationships marginal.
Cau tion should be taken in interpre ting these results. Bo rdin 1994)
contends that different treatments will produce different forms of alliance,
and presumably, different pattems of alliance development. It may be that
BRT therapists are taught to focus on rupture events while CBT therapists
are more likely to smooth over misunderstandings that may alter the
pattern of alliance development. While the fmdings of this study indicate
an area that should be investigated in further detail, there is not currently
enough information to draw any real conclusions.
It is possible that the way alliance and outcome were measured in this
study was not sensitive to the kind of changes that come from working
8/16/2019 Alliance Brief
18/22
AMERICAN JOURNA L O F PSYCH OTHER APY
within the session. Since the WAI can only look at shifts between sessions,
rather than within them, it is possible that it is missing a critical part of the
negotiation and change process. Researchers, such as Form an and M armar
(1985), Lansford (1986), and Safran, Muran and Samstag (1994), who
found that directly addressing strains, misunderstandings, and ruptures led
to improvement in alliance scores and predicted improved outcome,
looked at events that took place and we re address ed w ithin a given session.
The WAI may not measure the kinds of relational factors central to the
negotiating process. The WAI focuses on
greement
between patient and
therapist. It is clearly vital for patient and therapist to agree on what they
are trying to accomplish and how. This basic agreement on the central
goals of a treatment—and the methods used to achieve them—maybe a
fairly stable core of the therapeutic relationship. Shifts in this basic
agreem ent may reflect an inability to establish a firm foun dation for
treatment. What may be more important for some patients, however, is to
learn that they can dis gree with their therap ists without disrup ting the
relationship. The elements of the therapeutic relationship that may be
central to change may have much more to do with patients feeling their
therapist is willing to listen, understand and respect them, even when
expressing ang er, anxiety, disa ppo intm ent, hopelessness, etc. It may, there-
fore, be less important that the patient and therapist work through
conflicts about a particular task, than it is that the patient feel free to
express his or her doubt, discomfort, or skepticism and believe that the
therapist is taking him or her seriously. In these instances, a therapist's
willingness to explain a task, or change it, may be the curative factor.
Negotiations can take place in very brief moments, and may never be
explicitly stated, and thus, could not be reflected by an instrument that
measures problems lasting more than one session.
Similarly, the outcome measures, which reflect standard measures of
symptomatology and interpersonal functioning, may not reflect the kind of
changes that would be brought about by working through alliance rup-
tures. Safran and Muran (2000) suggest that working through ruptures,
w hethe r explicitly or implicitly, . . . can have an impact on the fundamen-
tal way in which patients con strue self-o the r intera ctions (p. 16). M uran
8/16/2019 Alliance Brief
19/22
Alliance Level and Pattem
treatment, but provides opportunities for a unique kind of growth that is
not reflected by standard measures.
Results from this study seem to suggest that the level of the alliance, as
measured by the WAI, rather than the shape of its development, is
predictive of outcome. Given the limitations of the WAI in tracking
interpersonal process, however, it may be more a function of the way in
which alliance is defined. The WAI has clearly demonstrated that the
alliance, defined by agreement on tasks, goals and bond, is a powerful
component of change in therapy. It may be necessary to develop other
tools to determine, with greater precision, how that change takes place.
Acknowledgments: The research presented in this article was supported in pan by Grsuit
MH50246 from the National Institute of Mental Health.
REFERENCES
Alexander, L.B.. & Luborsky, L. (1986). The Penn Helping Alliance Scales. In G reen berg , L.S., &
insof
W.M. (Eds.),
The psychotherapeutic process: A research handbook
(pp. 325-366). New
York: The Guilford Press.
American Psychiatric Association (1994). Diagnostic Systems Manual Fourth Edition. Washington:
Author.
Beck A.T., Freeman, A.. & Associates (1990). Cognitive therapy of personality disorders. New York;
Guilford.
Beck, J.S. (1995). Cognitive therapy: asics and heyond. New York: Guilford Press.
Bernard, H.S., Schwartz, A.J., Ocaltis, K.A,, & Stincr, A. (1980). The relationship between patients'
in-process evaluations of therapy and psychotherapy outcome. Journal of Clinical Psychology
36 259-2664 .
Bordin, E. (1979). The generalizability of the psychoanalytic concept of the working alliance.
Psychotherapy: Theory Research and Practice 16 ( 3 ) , 2 5 2 - 2 6 0 .
Bordin, E. (1994). Theory and research on the therapeutic working alliance: New directions. In
Horvath, A.O., Greenberg. L.S., (Ed.), The Working Alliance: Theory research and practice (pp.
13-37). New York: Joh n Wiley & Sons.
Derogatis, L.R. (1983). The Symptom Checklist -90 Revised: Adm inistration scoring and procedures
manual U. Baltimore: Clinical Psychometric Research.
Derogatis. L.R., & Lazarus, L. (1994). SCL-90—R, Brief Symptom Inventory, and matching ciinica
scales. In Maruish, M.E. (Ed.), The use of psychological testing for treatment planning and
outcome assessment (pp. 217-248). Hillsdale. NJ: Lawrence, Erlbaum.
Derogatis, L.R., & Spencer, P.M. (1982), Brief Symptom Inventory M anual. Baltimore: John Hop kins
School of Medicine.
DeRubies RJ., & Feeley, M. (1990). Determinants of change in cognitive therapy for depression.
Cognitive Therapy and Research 14. 4 6 9 - 4 8 2 .
En dico tt, J., Spitzer, R., Fleiss, J., & Coh en, J. (1976). Th e G lobal Assessment S cale. Achieves of
General Psychiatry H 166-171.
First, M.B., Gibbon. M., Spitzer R.L., Williams, J.B.W., & Benjamin, L. (1997).
Structured Clinical
Interview for DSM-AV Axis 1 1 Personality Disorders. W ashington, D C.: American Psychiatric
Publishing.
8/16/2019 Alliance Brief
20/22
8/16/2019 Alliance Brief
21/22
Alliance Level and Pattern
Horvaih (Ed.), The working alliance: Theo ry, Research and Practice (pp. 131-152). New York:
John Wiley Sons.
Rhodes, R.H., Hill, ;.E., Th om pso n, B.J., Elliott, R. (1994). Client ret ro sp ea ive recall of resolved
and unresolved misunderstanding events. Journal of ounseling Psychology, 41 (4), 473 -483 .
Safran. J.D. (2002). Brief Relational Psychoanalytic Treatment. Psychoanalytic Dialogues, 12 (2),
171-195.
Safran, J.D ., Cro cker , P., McM ain, S.. Mu rray, P. (1990). Th erap euti c alliance rup ture as a therapy
event for empirical investigation. Psychotherapy, 27, 154-165.
Safran, J.D., Mu ran, J.C. (1996). Th e resolution of ruptu res in the therapeu tic alliance.
Journal of
Consulting Clinical Psychology, 64, 4 4 7 -4 5 8 .
Safran, J.D ., M uran, J.C . (2000). Negotiating the therapeutic alliance: A relational treatment guide.
New York; Guildford Press.
Safran, J.D ., M uran , J,C ., Samstag, L.W . (1994). Resolving ther ape utic alliance rup ture s: A
task-analytic investigation. In A,O . Ho rvath L.S. Gree nberg (Eds.), The Working AUiance:
Theory, research and practice (pp. 225 -255 ). New York: John Wiley Sons.
Safran, J.D ., M uran , J . C . Samstag, L.W ,, Stevens, C. (2002), Repairing alliance rup ture s. In J.C .
Norcross (Ed.), Psychotherapy relationships that work (pp . 235-2 54). New York: Oxford Univereity.
Schwartz, A.J., Bernard, H.S. (1981). Com parison of patient and therapist evaluations of T im e-
Limited Psychotherapy. Psychotherapy: Theory, research and practice. 18, 101-107.
Tracey, T.J., Kokotovic, A.M., (1989). Factor structure of the W orking Alliance Inventory.
Psychological Assessment, 1989 (1), 207-210.
Turn er, A.E., M ura n,J.C . (1992). Cognitive-behavioral therapy for personality disorders: A treatment
manual. San Diag o. CA: Social Behavioral Do cum ents.
W inston, A., Laikin, M., McC ullough (1988).
Short-term dynam ic
therapy
for personality disorders:
A treatment manual. San Diego, CA: Social and Behavioral Documents.
8/16/2019 Alliance Brief
22/22
Top Related