Sei sulla pagina 1di 11

Copyright 1995 by the American Psychological Association, Inc.


Journal of Counseling Psychology

1995, Vol. 42, No. 3, 307-317

Attachment Patterns in the Psychotherapy Relationship: Development of

the Client Attachment to Therapist Scale
Diana L. Gantt

Brent Mallinckrodt

Laurel Hill Center, Eugene, Oregon

University of Oregon

Helen M. Coble
The Child Center, Springfield, Oregon
A panel of 9 therapists generated items for an instrument designed to measure the psychotherapy relationship from the perspective of attachment theory. The initial version of the
Client Attachment to Therapist Scale (CATS) contained 100 items that were administered at
4 counseling agencies in survey packets to 138 clients who had completed at least 5 sessions
with their therapists. Factor analysis suggested that 36 items loaded on 3 subscales, which we
labeled Secure, kvoidant-Fearful, and Preoccupied-Merger. CATS factors correlated in
expected directions with survey measures of object relations, client-rated working alliance,
social self-efficacy, and adult attachment. Cluster analysis identified 4 types of client
attachment. Significant differences in social competencies (object relations, etc.) were evident
across types of attachment. Implications of attachment patterns for the understanding of client
transference are discussed.

Attachment theory was originally developed to explain

the behavioral and emotional responses that keep young
children and their caregivers in close physical proximity
(Bowlby, 1969). From this foundation, attachment theory
offers an explanation for responses to separation and loss
(Bowlby, 1973) and the development of emotional attachments after infancy (Ainsworth, 1989; Bowlby, 1977). In an
optimal attachment bond, the caregiver provides a comforting presence for the child that reduces anxiety and promotes
feelings of security. From this secure base, the child is able
to explore the physical and social environment (Bowlby,
1969, 1988).
In Ainsworth's pioneering studies of attachment (Ainsworth, Blehar, Waters, & Wall, 1978), in sequential phases
of a laboratory observation, children explored a novel play
environment in the presence of their mother and were then
observed during a brief separation from their mother, exposure to a stranger, and reunion with their mother. Three
patterns of attachment were identified. Infants who displayed the secure pattern freely explored in their mother's

presence, showed some anxiety upon separation, and were

easily comforted upon reunion. Infants with the anxiousambivalent pattern were excessively anxious, angry, and
clinging to an extent that interfered with exploration, distressed during separation, and difficult to comfort upon
reunion. Anxious-avoidant infants showed little interest in
their mother and little strong affect throughout the observation. These attachment patterns have been confirmed in
subsequent studies including in-home observations (Egeland & Farber, 1984; Egeland & Sroufe, 1981) and are
thought to result from differences in consistency of caregivers' responses to the infant's physical and emotional
needs. Secure attachment may be promoted by caregivers
who are generally responsive. Ambivalent attachments may
develop when caregivers respond inconsistently, and avoidant attachments may develop when caregivers are consistently unresponsive and emotionally unavailable.
Early attachment experiences are thought to become
internalized, affecting the development of the child's concept of self and expectations about others (Bowlby, 1977;
Bretherton, 1985). A young child's working model of self
involves beliefs about whether she or he is generally the sort
of person worthy of care and deserving of help from others
versus being unworthy of help and comfort. The child's
working model of others involves generalized expectations
that caregivers will be responsive, helpful, and nurturing
versus unresponsive, aloof, and possibly harmful. The term
working models is used because in early childhood internal
representations can be revised as new attachments are encountered. However, working models become increasingly
resistant to change as development proceeds, because new
information that does not fit into existing structures is
difficult to process and tends to be defensively excluded
(Bowlby 1969, 1973; Bretherton, 1985). Beliefs about self

Brent Mallinckrodt, Counseling Psychology Program, University of Oregon; Diana L. Gantt, Laurel Hill Center, Eugene,
Oregon; Helen M. Coble, The Child Center, Springfield, Oregon.
Diana L. Gantt is now at the Providence Medical Center, Portland, Oregon.
We gratefully acknowledge cooperation of the staff and clients
of the DeBusk Center, the University of Oregon Counseling
Center, the Lane Community College Counseling Center, and the
Eugene Clinic. We also thank the students at the University of
WisconsinMadison and the students, staff, and faculty at the
University of MissouriColumbia who made helpful suggestions
about the interpretation of our results.
Correspondence concerning this article should be addressed to
Brent Mallinckrodt, Counseling Psychology Program, College of
Education, University of Oregon, Eugene, Oregon 97403.



and others formed through early attachment experience become an increasingly dominant part of a child's interpersonal style as the child brings working models to bear on
new relationships (Bowlby, 1988).
The attachment system established in childhood continues
to have a major influence on adult social relationships (for
reviews, see Coble, Gantt, & Mallinckrodt, in press; Hazan
& Shaver, 1994) and may be activated by any close, intimate relationship that evokes the potential for love, security,
and comfort, including friendship, kinship, romantic partnership, and the therapeutic alliance (Ainsworth, 1989).
Bowlby (1988) argued that the psychotherapy relationship
contains many features which activate an adult client's
ingrained attachment expectations and behaviors. Similar to
a parent or caregiver, the therapist offers emotional availability, a comforting presence, affect regulation, and a secure base from which to explore inner and outer worlds
(Pistole, 1989). Thus, the developing therapeutic relationship may be seen as a specialized form of adult attachment,
which is strongly influenced by a client's childhood attachment experiences (Bowlby, 1988). Expectations about the
relationship are influenced by the same working models of
self and others that a client applies to all close personal

of clients' attachment to their therapists. Components of

existing measuresfor example, the Bond subscale of the
Working Alliance Inventory (WAI; Horvath & Greenberg,
1989)no doubt capture an important part of this attachment relationship. However, because the development of
these measures was not explicitly guided by attachment
theory, important components may be missing from the
formulation. Therefore, the purpose of this study was to
develop and validate a measure that assesses client feelings
and attitudes toward the counselor from an attachment perspective. Clients' capacity to form positive, secure attachments to their therapists should be related to clients' capacity for healthy object relations (Ainsworth, 1969; Diamond
& Blatt, 1994), clients' capacity for attachment in adulthood
(Bowlby, 1988), the quality of the working alliance (Henry
& Strupp, 1994; Mallinckrodt, 1991), and the level of basic
social competencies (Mallinckrodt, 1992; Mallinckrodt,
Coble, & Gantt, 1995). Accordingly, we expected that a
valid measure of clients' attachment to their therapists
would be significantly correlated with measures of these

Thus, attachment theory concepts can provide a useful

new perspective on psychotherapy process. In therapy, the
client reexperiences a primary attachment, reproducing with
the therapist parts of an old and usually unsatisfactory
relationship (Jones, 1983). When a client reenacts attachment patterns in the therapeutic relationship, the therapist
gains access to the client's working models. As the therapist
and client explore their relationship and the client's relationships outside therapy, the client's working models become conscious and subject to challenge and change (Sperling & Lyons, 1994). The therapist acts as an empathic and
emotionally available attachment figure, encouraging an
examination of chronic dysfunctional patterns in relating to
others. In successful cases, a secure attachment relationship
is eventually formed with the therapist, contributing to a
corrective emotional experience for the client (Jones, 1983).
Recently, researchers have undertaken new applications
of attachment theory to understand psychotherapy processes. Bartholomew and Horowitz (1991) developed a
four-category model of adult attachment styles derived from
combinations of positive and negative working models of
self and others. In subsequent research, clients with different attachment styles were found to exhibit different patterns of interpersonal problems. These problem areas, in
turn, were found to differ markedly in how amenable they
were to change through brief dynamic psychotherapy
(Horowitz, Rosenberg, & Bartholomew, 1993). Client
memories of their early emotional bonds with parents have
also been found to be significantly associated with client
ratings of the working alliance (Mallinckrodt, 1991;
Mallinckrodt, Coble, & Gantt, 1995).
Although these studies have used concepts from attachment theory, we could locate no previous research that
conceptualized the therapeutic relationship from an attachment perspective and that attempted to measure the quality


Instrument Development
Nine experienced therapists (3 men and 6 women, all EuroAmerican in ethnicity) generated items in the initial development
of the Client Attachment to Therapist Scale (CATS). Three were
predoctoral interns, and six were psychologists with doctoral degrees with an average of 11.83 years of postgraduate experience
(range = 4-39). Participants were staff members of three university counseling centers or faculty members at two counseling
psychology programs. Participants completed materials individually at their own convenience. Packets contained a description of
the behavior displayed by infants of the attachment types described
by Ainsworth and her colleagues (1978), who used the "strange
situation" protocol. Participants were then told,
Research suggests these patterns may be relatively enduring
and may determine, for example, adult patterns of attachment
in romantic relationships. We hope to pilot test a measure
which would be able to detect patterns of secure, ambivalent,
and avoidant attachments of clients to their counselors.
Therapists were then directed to generate items for this measure
that were statements of opinion to which a client could respond
using a Likert-type scale anchored by strongly disagree through
strongly agree.
The panel generated a total of 143 items, which were pooled
with 129 items we had previously developed. Redundant items
were removed from the combined pool of 272 items, 8 new items
were written, and the wording of approximately 25 items was
changed to the negative to minimize response set bias. The resulting initial version of the CATS contained 100 Likert-type items
that used a 6-point response scale (see note to Table 1). After
pretesting with a panel of eight graduate students to reword unclear
items, we began data collection.


Instrument Validation
Clients were solicited for participation during a 3-year period
from four counseling agencies, including a university counseling
center, a community college counseling center, a hospital-based
outpatient clinic, and an in-house training clinic operated by a
counseling psychology program accredited by the American Psychological Association. All of the agencies were located in the
same community in the Pacific Northwest. Therapists were senior
staff, interns, or graduate students in training at the agencies.
Completed surveys were received from 138 clients. Of these, 62
(45%) were from the training clinic, 49 (35%) from the university
counseling center, 18 (13%) from the hospital clinic, and 9 (6%)
from the community college counseling center. The training clinic
participated in all 3 years of data collection, the university counseling center participated for 2 years, and both the hospital clinic
and community college counseling center participated for only 1
year. Few clients at the training clinic or hospital clinic were
college students. Thus, approximately half of the sample were
community residents. Both clinics accepted clients for relatively
long-term treatment, whereas both counseling centers operated
primarily from a brief therapy model. Part of the data from 77 of
these clients was reported in a previous publication (Mallinckrodt,
Coble, & Gantt, 1995).
Our procedures did not allow for a determination of the survey
compliance rate, but the 62 clients participating from the training
clinic represented 52% of all clients seen for more than five
sessions during the data collection period. At the point they were
surveyed, clients had completed a median of 10 sessions with their
current therapist (range = 5-62). Of the 138 clients, 121 (88%)
were women, 15 (11%) were men, and 2 (1%) did not indicate
their sex. In 102 dyads, the counselor's sex was known, with 75
(74%) of therapists being women and 27 (26%) men. Mean age of
the clients was 32.57 years (SD = 10.86, range = 18-64).
Regarding ethnic identification, 122 (88%) of the clients indicated
Caucasian, 2 (1%) Hispanic, 2 (1%) Native American, and 12
(10%) indicated "other" or left the item blank.

In addition to the CATS items, survey packets contained the
measures described below. However, the Bell Object Relations
and Reality Testing Inventory (BORRTI) was included only during the last 2 years of the data collection period. Thus, only 72 of
the 138 clients completed all of the survey measures.
Working Alliance Inventory. The WAI (Horvath & Greenberg,
1986, 1989) is a 36-item self-report measure that uses a 7-point
fully anchored response scale (1 = never, 7 = always). Only the
36-item client form was used in this study. The WAI consists of
three subscales to assess (a) the emotional bond of trust and
attachment between counselor and client, (b) the agreement concerning the overall goals of treatment, and (c) the agreement
concerning the tasks relevant for achieving these goals. Good
construct validity has been established through multitraitmultimethod analyses (Horvath & Greenberg, 1986).
Bell Object Relations and Reality Testing Inventory. The BORRTI (Bell, 1991; Bell, Billington, & Becker, 1986) is a self-report
measure of ego functioning and object relations. Only the 45
true-false items of the object-relations domain were used in this
study. Four subscales measure the following object-relations deficits: Alienation, a lack of basic trust in relationships, inability to
attain closeness, and hopelessness about achieving stable and


satisfying levels of intimacy; Insecure Attachment, painfulness in

interpersonal relations, sensitivity to rejection, and excessive concerns about being liked and accepted; Egocentricity, mistrust of the
motivations of others, others existing only in relation to oneself,
and others being manipulated for one's own aims; and, Social
Incompetence, shyness, uncertainty about how to interact with the
opposite sex, and inability to make friends. Bell (1991) reported
high factorial stability for these four subscales, and internal consistency and split-half reliabilities that ranged from .78 to .90. The
BORRTI has demonstrated significant known-groups discriminant
validity in tests that compared patients with borderline personality
disorder, major affective disorder, schizophrenia, and adult volunteers (Bell, 1991).
Adult Attachment Scale (AAS). The AAS (Collins & Read,
1990) consists of 18 items scored on a 5-point scale that ranges
from not at all characteristic of me (1) to very characteristic of me
(5). Factor analysis identified three subscales of six items each.
The Depend subscale measures the extent participants trust others
and rely on them to be available if needed. The Close subscale
assesses comfort with intimacy and emotional closeness. The
Anxiety subscale measures fears of abandonment. Internal consistency reliability (coefficient alpha) and retest reliability after a
2-month interval were greater than .58 for the three subscales
(Collins & Read, 1990). Subscale scores were correlated in theoretically expected directions with measures of self-esteem, social
behavior, instrumentality, expressiveness, openness, and satisfaction in romantic relationships (Collins & Read, 1990).
Self-Efficacy Scale (SES). The SES (Sherer et al., 1982) was
selected because it yields a separate measure of "social" selfefficacy that has been suggested as an important component of the
ability to establish healthy attachments in adulthood (Coble, Gantt,
& Mallinckrodt, in press). Responses to the 23 items of the original
SES are made on a 14-point scale, but in the current study a 5-point
response format was used (1 = strongly disagree, 5 = strongly
agree). Factor analyses identified 17 items that load on the General
Self-Efficacy subscale and 6 items that load on the Social SelfEfficacy subscale. Internal consistency (coefficient alpha) of .86
and .71 were obtained for the General Self-Efficacy and the Social
Self-Efficacy subscales, respectively. Evidence of construct validity was provided by significant correlations of the General SelfEfficacy and the Social Self-Efficacy subscales in the predicted
direction with internal locus of control, interpersonal competency
assertiveness, and the Minnesota Multiphasic Personality Inventory Social Introversion Scale 0 (Sherer & Adams, 1983; Sherer et
al., 1982).

Announcements posted in the agencies described the study as a
survey of how "emotional bonds formed between persons in childhood with their parents affect preferences for types of relationships
as adults, and the types of counseling relationships they develop."
As an incentive, participants received a gift certificate worth $3 for
a videotape movie rental or movie ticket. Because we felt a
minimum amount of contact was necessary for important features
of the client-therapist attachment relationship to emerge, clients
were asked to participate only after they had completed at least
five sessions of individual counseling. To increase variability,
clients were allowed to participate no matter how many sessions
they had completed beyond the fifth session when data collection
began at their counseling agency. Participants completed surveys
in their homes and returned gift certificate requests and survey
materials in separate prestamped envelopes. Clients were promised
complete anonymity and did not label survey materials with any



personally identifying information. In the last year of data collection, a subsample of 17 consecutive clients who chose to participate at one of the data collection sites provided retest data by
completing a second version of the 100-item CATS instrument
2-4 weeks after the initial survey.


Factor Analysis and Item Selection

Our first task was to reduce the initial pool of 100 items
to a more manageable number and investigate the factor
structure of the data. Before we proceeded with a factor
analysis, the condition of the data matrix was examined by
calculating the Kaiser-Meyer-Olkin (KMO) measure of
sampling adequacy. This measure indicates the amount of
shared variance in the item pool, and may range from zero
to one. The initial KMO index for the 100 X 100 matrix was
.288, a level considered unacceptable and characteristic of
matrices in which some items share little variance with
other items in the pool (Kaiser, 1974). This result was not
surprising, given our emphasis on creating a diverse pool of
items. Output produced by the SPSSX software package
indicates items with low "anti-image correlations," that is,
those items that are essentially uncorrelated with other items
in the pool (Norusis, 1985). Items with the lowest antiimage correlations were removed in stepwise fashion one at
a time, until the item pool was reduced to 75 items. Many of
the 25 items removed seemed conceptually less related to
attachment, generally, than the items that were retained. The
KMO index for the 75-item pool was .749, a level considered acceptable (Kaiser, 1974). Bartlett's test for sphericity
was also significant, ^ ( 2 , 775, N = 75) = 7829.87, p <
.001, indicating that the 75-item matrix was significantly
different from a matrix of essentially unrelated items.
A ratio of 5-10 participants per item, up to a maximum of
300 participants in total, has been recommended as a minimum acceptable sample size for factor analysis (Tinsley &
Tinsley, 1987). The ratio of 1.84 participants per item in this
data set falls below this standard. However, an investigation
of the effects of sample size demonstrated that virtually the
same factor structure resulted from a sample of 14 participants per item (on a 76-item scale) as was obtained from a
subsample of 1.3 participants per item (Arrindell & van der
Ende, 1985). We decided to proceed with a factor analysis
using our sample of 138 clients, recognizing that findings
must be considered tentative until replicated with larger
A principal-factors analysis using squared multiple correlations for commonality estimates, with oblique rotation
to extract nonorthogonal factors, was performed on the 75
items remaining in the pool. A scree test (Cattell, 1966)
suggested that three factors with eigenvalues greater than
one should be retained. A second principal-factors analysis
with oblique rotation was performed with a forced threefactor solution. The resulting three factors accounted for
26%, 7%, and 5% of variance in the data.
Items loading greater than .40 on a factor were used to
construct the three initial subscales of the CATS, which

contained 29, 24, and 11 items, respectively. The factors

seemed readily interpretable as extracted, but our goal, if
possible, was to hold the instrument to fewer than 40 items
to increase ease of administration. Therefore, items were
removed from the three initial subscales if (a) the item
seemed conceptually unrelated to other items on the subscale and its removal increased internal consistency (coefficient alpha) of the subscale or (b) the item was highly
correlated or seemed redundant with another item on the
same subscale. After this screening, a total of 36 items were
retained, none of which loaded on more than one subscale
greater than .40. The final version of the CATS instrument,
together with item loadings, is presented in Table 1. The
labels and interpretations of the three subscales are as follows: (a) Secure (14 items), experiencing the therapist as
responsive, sensitive, understanding, and emotionally available; feeling hopeful and comforted by the counselor; and
feeling encouraged to explore frightening or troubling
events; (b) Avoidant-Fearful (12 items), suspicion that the
therapist is disapproving, dishonest, and likely to be rejecting if displeased; reluctance to make personal disclosures in
therapy; and feeling threatened, shameful, and humiliated in
the sessions; (c) Preoccupied-Merger (10 items), longing
for more contact and to be "at one" with the therapist,
wishing to expand the relationship beyond the bounds of
therapy, and preoccupation with the therapist and the therapist's other clients.
The stability of the factor structure was investigated by
repeating the analysis with principal-components extraction. When items that loaded greater than .40 were examined, the set was identical to that obtained from the principal-factors extraction. Next, the sample was divided into
two randomly selected halves. A factor analysis with principal-factors extraction and oblique rotation was performed
independently on each half of the sample. Unfortunately,
these analyses could not be performed on the complete set
of 75 items because half the sample (n = 69) contained
fewer participants than items. Instead, the 36 items selected
for the final version of the CATS were used. In the first
analysis, 2 of the 36 items loaded on factors different than
the original analysis with the full sample. In the second
split-half analysis, 3 of the 36 items loaded on different
factors. Despite the shifting allegiance of some items in the
split-half analyses, the factor loadings of the original analysis from the full sample were retained.

Psychometric Properties of the CATS

Means and standard deviations for the CATS subscales
are contained in Table 2, together with internal consistency
and retest reliabilities. The 17 clients who provided retest
data completed the second CATS survey an average of 3.24
weeks after completing the first instrument (range = 2.05.14). Internal consistency (coefficient alpha) and retest
reliability coefficients (Pearson's product-moment correlations) were greater than .63 for all of the subscales. Regarding subscale correlations, the Avoidant-Fearful and Secure
subscales were significantly negatively correlated, whereas



Table 1

Client Attachment to Therapist Scale (CATS) Items and Subscale Factor Loadings
23 a


Item text
Factor 1: Secure (14 items)
I don't get enough emotional support from my counselor.
My counselor is sensitive to my needs.
My counselor is dependable.
I feel that somehow things will work out OK for me when I am
with my counselor.
My counselor isn't giving me enough attention.
When I show my feelings, my counselor responds in a helpful
I don't know how to expect my counselor to react from session
to session.
I can tell that my counselor enjoys working with me.
I resent having to handle problems on my own when my
counselor could be more helpful.
My counselor helps me to look closely at the frightening or
troubling things that have happened to me.
My counselor is a comforting presence to me when I am upset.
I know my counselor will understand the things that bother me.
I feel sure that my counselor will be there if I really need her/
When I'm with my counselor, I feel I am his/her highest
Factor 2: Avoidant/fearful (12 items)
I think my counselor disapproves of me.
Talking over my problems with my counselor makes me feel
ashamed or foolish.
I know I could tell my counselor anything and s/he would not
reject me.
I don't like to share my feelings with my counselor.
I feel humiliated in my counseling sessions.
Sometimes I'm afraid that if I don't please my counselor, s/he
will reject me.
I suspect my counselor probably isn't honest with me.
My counselor wants to know more about me than I am
comfortable talking about.
I feel safe with my counselor.
My counselor treats me more like a child than an adult.
It's hard for me to trust my counselor.
I'm not certain that my counselor is all that concerned about


- .59

Factor 3: Preoccupied/merger (10 items)

I yearn to be "at one" with my counselor.
I wish my counselor could be with me on a daily basis.
I would like my counselor to feel closer to me.
I'd like to know more about my counselor as a person.
I think about calling my counselor at home.
I think about being my counselor's favorite client.
I wish there were a way I could spend more time with my
I wish I could do something for my counselor too.
I wish my counselor were not my counselor so that we could
be friends.
I often wonder about my counselor's other clients.
Note. N = 138. In the interest of promoting further study, other researchers may use this scale
without contacting us to obtain prior permission. However, we do ask that researchers send any
reports of research findings as soon as available, including those that remain unpublished, to Brent
Mallinckrodt. For comparability to the norms published in this study, the CATS should be prefaced
with these instructions: "These statements refer to how you currently feel about your counselor.
Please try to respond to every item using the scale below to indicate how much you agree or disagree
with each statement." Instructions should be followed with this response scale: 1 = strongly
disagree, 2 = somewhat disagree, 3 = slightly disagree, 4 = slightly agree, 5 = somewhat agree,
6 = strongly agree. To score the CATS, reverse key (i.e., 6 = 1, 5 = 2, etc.) the six items with
negative subscale loadings, then sum the items for each subscale. This procedure will result in higher
scores indicating more Secure, Avoidant-Fearful, and Preoccupied-Merger attachments.
Item should be reverse keyed.



Table 2
Psychometric Properties of the Client Attachment to Therapist Scale (CATS) Subscales




of items
Skewa (statistical significance)
Internal consistency
(coefficient alpha)
Test-retest reliability

5-8 sessions' (n = 45)
9-15 sessions'1 (n = 44)
16-62 sessions'" (n = 45)
Note. N = 134 because of missing data for some CATS subscales.
Statistical skew represents deviation from a normal distribution. Negative skew indicates a larger
than expected number of high scores. Positive skew indicates an abundance of low scores.
Norms by length of therapy at time of data collection.
Mean is significantly different than the means for the other two groups of client responses for this
**p < .01.

the Secure and Preoccupied-Merger subscales were positively correlated.

Because clients varied considerably with regard to the
number of sessions they had completed at the time of data
collection, the sample was divided into three groups of
approximately equal size on the basis of length of therapy.
Table 2 contains the results of analyses of variance
(ANOVAs) used to compare means of the three CATS
subscales for these three groupings. Comparisons suggested
that the Secure subscale scores were significantly different
depending on length of therapy, F(2, 135) = 4.26, p < .05.
Duncan's multiple-range test, used for follow-up group
comparisons, indicated that clients seen for 5-8 sessions at
the time of data collection had significantly lower Secure
subscale scores than did either of the other two groups seen
for a longer period. Perhaps a secure attachment takes
relatively longer to develop, or clients with low levels of
secure attachment tend not to remain in therapy beyond
eight sessions. In any case, these findings suggest that other
researchers using the CATS for comparative purposes
should use the most appropriate group norms reported in
Table 2.
Evidence of concurrent validity is reported in Table 3.
The pattern of correlations corresponds closely to expectations based on our empirical definitions of the constructs
measured by the CATS subscales. Clients who perceived
their therapeutic attachment as secure (i.e., high scores on

CATS Secure subscale) were relatively free of object-relations deficits and tended to report positive working alliances. In contrast, clients with high scores on the CATS
Avoidant-Fearful subscale, as expected, tended to have
much less positive working alliances and broad deficits in
object relations. Clients who scored relatively high on the
third CATS subscale, Preoccupied-Merger, tended to exhibit object-relations deficits in insecure attachment. It is
interesting that these clients tended to rate the bond aspects
of their working alliance as positive, but not the task and
goal aspects. As expected, the Avoidant-Fearful and the
Preoccupied-Merger subscales were negatively correlated
with General Self-Efficacy. It is interesting that AvoidantFearful scores, but not Preoccupied-Merger scores, were
negatively associated with ratings of self-efficacy for social
outcomes. Contrary to expectations, the Secure subscale
was not significantly associated with self-efficacy. Correlations with the AAS subscales were less consistently supportive of CATS construct validity.
Finally, we conducted a cluster analysis to identify patterns of clients' attachment to their therapists. We used four
measures of the therapeutic relationship (the three CATS
subscales plus client-rated working alliance) as grouping
variables. To eliminate scaling differences, grouping variables were standardized. Cluster analysis begins by calculating a proximity matrix of the squared-Euclidean distance
between all pairs of clients for the set of sorting variables.



Table 3
Correlations of the Client Attachment to Therapist Scale Subscales With Other

Working Alliance
Adult Attachment
Self-Efficacy Scale
General Self138
Social SelfEfficacy
Note. Sample sizes vary because of missing
Testing Inventory.
*p<.05. **p<.01.





























data. BORRTI = Bell Object Relations and Reality

We used Ward's (1963) method of forming clusters, which

sequentially combines pairs of individuals or clusters
formed at an earlier step while seeking to minimize withingroup variability and maximize between-group variability
in Euclidean distance. At each step, the within-group variance is monitored. A sharp increase in variance at a given
step indicates that the number of clusters formed in the
previous step provides the best fitting description of the data
(Borgen & Barnett, 1987). This procedure suggested that a
four-cluster solution provided the best fit for our data, with
23, 27, 56, and 20 clients assigned to the four clusters.
(Because of missing data, only 126 clients were classified.)
Because we wanted to form clusters that were exclusively
based on characteristics of the counseling relationship, and
to use as many clients as possible in the analysis, only the
four measures of the counseling relationship were used as
grouping variables. However, in order to describe and differentiate the clusters, we examined counseling relationship
variables plus measures of object relations, adult attachment, and self-efficacy. We termed this second set of nine
subscales social competencies because they tap a client's
capacity for forming healthy intimate relationships. Complete data for social competencies measures were available
from only 68 clients. (Recall that the BORRTI measure of
object relations was introduced in the 2nd year of data
collection.) Comparisons indicated that the 68 clients with
complete data did not differ from the 58 clients with missing
social competencies data regarding any of the variables used
to form the clusters, fs(124) < 1.21, ps > .23. Thus, to

include social competencies measures, we based subsequent

examination of the clusters on the 68 clients with complete
Results of a multivariate analysis of variance used to
compare the four clusters indicated significant differences
in measures of therapeutic relationship and social competencies, F(39, 162) = 1.96, p < .001. Univariate follow-up
analyses indicated significant differences Fs(3, 64) > 3.81,
ps < .01, in all of the measures except for the Close and
Anxiety subscales of the AAS and the Social Self-Efficacy
subscale. Figure 1 provides a graphic depiction of these
differences. Members of the largest cluster (n = 33), which
we labeled secure, had positive working alliances and high
scores on the CATS Secure subscale. The second, and
smallest, cluster, labeled reluctant (n = 6), was similar to
the first cluster in scores on the CATS Secure subscale, but
scores on the Avoidant-Fearful subscale were much higher
for the reluctant cluster than for the secure cluster. The third
cluster, labeled avoidant (n = 11), had high scores on the
CATS Avoidant-Fearful subscaleas did the reluctant
clusterbut unlike the other three clusters, working alliance ratings in the avoidant cluster were very poor. The
fourth cluster, labeled merger (n 18), was distinguished
by high scores on the CATS Preoccupied-Merger subscale.
The second panel of Figure 1 illustrates the marked object-relations deficits and lower levels of general self-efficacy, which are characteristic of the avoidant and merger
clusters and a high willingness to depend on others in adult
attachments. It is interesting that although both clusters




E 0.8





15 0.4




(n = 33)

CATS Subscales








(n = 6)




(n = 11)




Figure 1. Profiles of four Client Attachment to Therapist Scale (CATS) clusters regarding
therapeutic relationship variables (A) and social-competencies variables (B). Object-relation deficits
are measured by the Insecure Attachment (InA), Alienation (Aln), Egocentricity (Ego), and Social
Incompetence (Scl) subscales of the Bell Object Relations and Reality Testing Inventory. Adult
attachment is measured by the Close, Depend (Dep), and Anxiety (Anx) subscales of the Adult
Attachment Scale. Self-efficacy is measured by the General Self-Efficacy (GSe) and the Social
Self-Efficacy (SSe) subscales of the Self-Efficacy Scale.

exhibited high Depend scores, members of the avoidant

cluster had high object-relations "social incompetence" deficits, whereas members of the merger cluster rated themselves as less socially incompetent. Those in the secure
cluster exhibited relatively lower levels of object-relations
deficits. Although the reluctant and avoidant clusters had
virtually identical scores on the CATS Avoidant-Fearful
subscale, the reluctant cluster had much lower levels of
object-relations deficits than did the avoidant cluster. Finally, an ANOVA used to test differences in the number of
sessions completed at data collection indicated no significant differences in clusters, F(3, 64) = 0.61, p = ns. The
mix of male and female clients in each cluster was also not
significantly different, ^ ( 3 , n = 66) = 5.98, p = ns.

The purpose of this study was to develop and validate a
measure that assessed the client-therapist relationship from
the perspective of attachment theory. A diverse group of
experienced therapists participated in developing the measure, and a diverse group of clients participated in validating
it. Factor analysis revealed one factor, which seemed to
capture client perceptions of secure attachment, that accounted for a sizable amount of variance and two smaller
factors, which both seemed to capture more troubled attachments to the therapist. The subscales derived from these
factors demonstrated acceptable internal and retest reliability. The overall pattern of correlations with other measures


supports the construct validity of the CATS. The low correlations between CATS subscales, and different patterns of
association with other measures, suggest that the subscales
measure distinct aspects of clients' attachment relationship
to their therapists. Because shared variance with the WAI
was relatively low for the second and third CATS subscales,
and lower than the variance shared by the WAI subscales
with one another, findings suggest that these CATS subscales measure an aspect of the counseling relationship
distinct from the working alliance.
Just as patterns of attachment have been observed in
infants (Ainsworth et al., 1978) and are believed to govern
adult attachment (Bartholomew & Horowitz, 1991), we
hoped to identify patterns of attachment in psychotherapy.
Results of the concurrent validity analyses and cluster analysis suggest that there are distinct patterns in clients' attachment to their therapists. Clients who scored high on the
CATS Secure subscale perceived their therapists as emotionally responsive, accepting, and promoting a "secure
base" (Bowlby, 1988) from which to explore threatening
aspects of their emotional experience. Clients classified in
the secure cluster tended to report positive working alliances, good object-relations capacity, and a relatively strong
sense of self-efficacy. These clients probably have a highly
positive working model of others and a fairly positive
working model of self.
Clients who scored high on the CATS PreoccupiedMerger subscale, many of whom were classified in the
merger cluster, seemed to desire a dissolution of normal
boundaries in the therapy relationship. Not only did they
wish more frequent and intensely personal contact, but these
clients literally wished to be "at one" with their therapists.
They also tended to be compulsively preoccupied with the
therapist and the therapist's other clients. Clients in the
merger cluster reported a high willingness to depend on
others and also a number of serious object-relations deficits.
The correlations of the CATS Preoccupied-Merger subscale
with working alliance dimensions suggest that these clients
form a working alliance bond with their therapists much
more readily than they come to agreement about the tasks or
goals of therapy. It seems likely that clients in the merger
cluster have maintained a strongly negative working model
of self and a positive working model of others.
Clients who scored high on the Avoidant-Fearful subscale tended to distrust their therapists and fear rejection.
They were reluctant to cooperate in the self-disclosure tasks
of therapy and tended to feel ashamed and humiliated during sessions. This subscale was correlated with objectrelations deficits involving egocentricity, social incompetence, and alienation (a lack of basic trust in relationships
and hopelessness about deriving satisfaction from them).
Clients in the avoidant cluster scored high on the CATS
Avoidant-Fearful subscale and reported the poorest working alliances, but they also exhibited high scores on the
AAS Close and Depend subscales and on the BORRTI
Social Incompetence subscale. This combination suggests
strong yearnings for emotional connection but grave questions about one's ability to establish such a relationship. It
seems doubtful that clients with an unambiguously negative


working model of others would stay voluntarily in therapy

for very long. Clients in the avoidant cluster may have a
strongly negative working model of self and a negative
model of others with some ambivalence. They may experience themselves as generally not likable and most others as
potentially dangerous and rejecting.
The small number of clients in the reluctant cluster are
more difficult to categorize. As with members of the
avoidant cluster, they scored high on the CATS AvoidantFearful subscale; but unlike the avoidant cluster, clients
classified as reluctant had few object-relations deficits
(lower scores than any cluster), relatively positive working
alliances, and high scores on the CATS Secure subscale. We
chose the label reluctant for this cluster to connote unwillingness to participate in the self-revealing tasks of therapy
(measured by the second CATS subscale). Despite this
possible hesitance and feelings of humiliation in sessions,
these clients nevertheless seemed engaged with their therapists. They reported their therapists as emotionally responsive and their working alliance as fairly strong. It is tempting to speculate that these therapists and clients had
somehow managed to "break through" to engage each other,
but much more research is needed on this point. An alternative view is that a considerable degree of denial influenced these clients' self-ratings of object-relations deficits
and therapy relationships.
Taxonomies of personality disorder involving categories
such as "compulsive care seeking" and "angry withdrawal"
have been developed on the basis of dysfunctional patterns
of attachment (West & Sheldon, 1990). Perhaps clients in
the merger or avoidant clusters in this study, that is, those
with elevated scores on the CATS Preoccupied-Merger or
Avoidant-Fearful subscales, manifest significant personality dysfunction. Clients classified in the reluctant cluster in
this study may have a relatively positive working model of
self, but a negative working model of others (perhaps excluding their therapists). If so, then the four clusters we
identifiedsecure, merger, avoidant, and reluctantcorrespond closely to the four types of adult attachment in
Bartholomew and Horowitz' (1991) 2 X 2 model, which
they termed secure, preoccupied, fearful, and dismissing.
Subsequent research suggested that clients with a dismissing attachment styleand thus perhaps those in our reluctant clustermay benefit least from brief psychotherapy
(Horowitz et al., 1993). Other research has suggested that
the experience of incestuous sexual abuse interferes with the
adult capacity to use social support and form intimate attachments (Mallinckrodt, McCreary, & Robertson, 1995).
Sexual abuse survivors among the clients in the current
study may have heightened difficulties with forming healthy
attachments to their therapists.
Working models of self and others, developed through
early attachment experiences, may form the basis for a
client's fundamental interpersonal orientation, which can be
described in terms of the circumplex dimensions of affiliation and dominance and in terms of patterns of adult attachment (Horowitz et al., 1993). Psychotherapeutic transference involves client feelings and perceptions that correctly
belong to earlier relationships that are displaced onto the



therapist (Gelso & Carter, 1994). Viewed from the perspective of attachment theory, transference may be understood
as a misperception of the therapist and of the therapeutic
relationship resulting from the client's use of long-established working models of self and others to resolve ambiguities in the new caregiving (therapeutic) attachment and
to anticipate the motives and behavior of the new attachment figure (therapist). Thus, to the extent that the CATS
measures misperception of the therapeutic relationship, subscale scores may be influenced by client transference. This
possibility is intriguing, but much more research is needed
to relate patterns of attachment with a therapist to general
patterns of adult attachment.
A number of important limitations in this study should be
noted. All measures were self-report, which introduces the
possibility that correlations are inflated and results biased
because of common method variance and response set.
More confidence could be placed in the results of the factor
analysis if a larger sample had been available. Findings need
to be confirmed in new samples. Use of the counselor's
ratings of the working alliance would have provided a
valuable piece of information not available in this data set.
Premature termination undoubtedly influenced our findings,
given the evidence that Secure attachment scores were
significantly higher in clients who had been seen for eight
sessions or more. The small number of male clients and
clients representing ethnic diversity limits the generalizability of the findings. Secure attachment between minority
clients and their therapists may fail to develop because of
clients' quite functional reaction to the therapists' lack of
multicultural sensitivity (Coleman, 1994). Although we
measured only characteristics of clients, therapists offer the
relationship conditions that make an attachment possible,
and it would be a serious mistake to attribute difficulty in
establishing a secure psychotherapy attachment entirely to
client factors. Finally, the CATS taps opinions that clients
might be very reluctant to share with their therapists. Clients
in this study completed the measure anonymously, a procedure that enhanced reliability of the data for research purposes but that would severely restrict the usefulness of the
CATS for clinical purposes in ongoing therapy.
Further research is needed to test the psychometric properties of the CATS with other samples of clients. If the
CATS is established as a valid and reliable measure, a
number of interesting avenues for further study would be
opened. Memories of childhood attachment experience associated with each pattern of attachment to therapist could
be identified. The dynamics of the unfolding therapy relationship that are characteristic of each type of attachment
could be investigated, together with variables such as suitability for brief therapy, anticipated breaches in the working
alliance (Safran & Muran, 1995), methods of establishing a
productive alliance, and the corrective emotional experiences that are most likely to lead to therapeutic change.
Perhaps the process of therapeutic change itself could be
tracked as changes over the course of therapy in a client's
attachment patterns and flexibility of working models.

Ainsworth, M. D. S. (1969). Object relations, dependency and
attachment: A theoretical view of the mother-infant relationship.
Child Development, 40, 969-1025.
Ainsworth, M. D. S. (1989). Attachments beyond infancy. American Psychologist, 44, 709-716.
Ainsworth, M. D. S., Blehar, M. C, Waters, E., & Wall, S. (1978).
Patterns of attachment: A psychological study of the strange
situation. Hillsdale, NJ: Erlbaum.
Arrindell, W. A., & van der Ende, J. (1985). An empirical test of
the utility of the observations-to-variables ratio in factor and
components analysis. Applied Psychological Measurement,
9, 165-178.
Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles
among young adults: A test of a four-category model. Journal of
Personality and Social Psychology, 61, 226-244.
Bell, M. D. (1991). An introduction to the Bell Object Relations
Reality Testing Inventory. Unpublished manuscript. (Available
from Morris Bell, Department of Psychiatry, Veterans Administration Hospital, West Spring Street, West Haven, CT 06516)
Bell, M., Billington, R., & Becker, B. (1986). A scale for the
assessment of object relations: Reliability, validity and factorial
invariance. Journal of Clinical Psychology, 42, 733-741.
Borgen, F. H., & Barnett, D. C. (1987). Applying cluster analysis
in counseling psychology research. Journal of Counseling Psychology, 34, 456-468.
Bowlby, J. (1969). Attachment and loss: Vol. 1. Attachment. New
York: Basic Books.
Bowlby, J. (1973). Attachment and loss: Vol. 2. Separation
Anxiety and anger. London: Hogarth Press.
Bowlby, J. (1977). The making and breaking of affectional bonds:
1. Aetiology and psychopathology in the light of attachment
theory. British Journal of Psychiatry, 130, 201-210.
Bowlby, J. (1988). Attachment, communication, and the therapeutic process. In A secure base: Parent-child attachment and
healthy human development (pp. 137-157). New York: Basic
Bretherton, I. (1985). Attachment theory: Retrospect and prospect.
Monographs for the Society for Research in Child Development,
50(1-2), 3-35.
Cattell, R. B. (1966). The scree test for the number of factors.
Multivariate Behavioral Research, 1, 245-276.
Coble, H. M., Gantt, D. L., & Mallinckrodt, B. (in press). Attachment, social competency, and the capacity to use social support.
In G. Pierce, B. R. Sarason, & I. G. Sarason (Eds.), Handbook
of social support and the family. New York: Plenum.
Coleman, H. L. K. (1994, April). Strategies for coping with cultural diversity. Paper presented at the annual conference of the
American Counseling Association, Minneapolis, MN.
Collins, N. L., & Read, S. J. (1990). Adult attachment, working
models, and relationship quality in dating couples. Journal of
Personality and Social Psychology, 58, 644-663.
Diamond, D., & Blatt, S. J. (1994). Internal working models and
the representational world in attachment and psychoanalytic
theories. In M. B. Sperling & W. H. Berman (Eds.), Attachment
in adults: Clinical and developmental perspectives (pp. 72-97).
New York: Guilford Press.
Egeland, B., & Farber, E. (1984). Infant-mother attachment: Factors related to its development and change over time. Child
Development, 55, 753-771.
Egeland, B., & Sroufe, L. A. (1981). Attachment and early maltreatment. Child Development, 52, 44-52.
Gelso, C. J., & Carter, J. A. (1994). Components of the psycho-


therapy relationship: Their interaction and unfolding during
treatment. Journal of Counseling Psychology, 41, 296-306.
Hazan, C, & Shaver, P. R. (1994). Attachment as an organizational framework for research on close relationships. Psychological Inquiry, 5, 1-27.
Henry, W. P., & Strupp, H. H. (1994). The therapeutic alliance as
interpersonal process. In A. O. Horvath & L. S. Greenberg
(Eds.), The working alliance: Theory, research and practice (pp.
51_84). New York: Wiley.
Horowitz, L. M., Rosenberg, S. E., & Bartholomew, K. (1993).
Interpersonal problems, attachment styles, and outcome in brief
dynamic psychotherapy. Journal of Consulting and Clinical
Psychology, 61, 549-560.
Horvath, A. O., & Greenberg, L. (1986). The development of the
Working Alliance Inventory. In L. Greenberg & W. Pinsoff
(Eds.), The psychotherapeutic process: A resource handbook
(pp. 529-556). New York: Guilford Press.
Horvath, A. O., & Greenberg, L. (1989). Development and validation of the Working Alliance Inventory. Journal of Counseling Psychology, 36, 223-232.
Jones, B. (1983). Healing factors of psychiatry in light of attachment theory. American Journal of Psychotherapy, 35, 235-244.
Kaiser, H. F. (1974). An index of factorial simplicity. Psychometrika, 39, 31-36.
Mallinckrodt, B. (1991). Clients' representations of childhood
emotional bonds with parents and formation of the working
alliance. Journal of Counseling Psychology, 38, 401-408.
Mallinckrodt, B. (1992). Childhood emotional bonds with parents,
development of adult social competencies, and the availability of
social support. Journal of Counseling Psychology, 39, 453-461.
Mallinckrodt, B., Coble, H. M., & Gantt, D. L. (1995). Working
alliance, attachment memories, and social competencies of
women in brief therapy. Journal of Counseling Psychology,
42, 79-84.
Mallinckrodt, B., McCreary, B. A., & Robertson, A. K. (1995).


Co-occurrence of eating disorders and incest: The role of attachment, family environment, and social competencies. Journal of
Counseling Psychology, 42, 178-186.
Norusis, M. J. (1985). SPSSX advanced statistics guide. New
York: McGraw-Hill.
Pistole, C. M. (1989). Attachment: Implications for counselors.
Journal of Counseling and Development, 68, 190-193.
Safran, J. D., & Muran, J. C. (1995). The resolution of ruptures in
the therapeutic alliance. Manuscript submitted for publication,
New School for Social Research, Beth Israel Medical Center.
Sherer, M., & Adams, C. H. (1983). Construct validation of the
Self-Efficacy Scale. Psychological Reports, 53, 899-902.
Sherer, M., Maddux, J. E., Mercadante, B., Prentice-Dunn, S.,
Jacobs, B., & Rogers, R. W. (1982). The Self-Efficacy Scale:
Construction and validation. Psychological Reports, 51,
Sperling, M. B., & Lyons, L. S. (1994). Representations of attachment and psychotherapeutic change. In M. B. Sperling & W. H.
Berman (Eds.), Attachment in adults: Clinical and developmental perspectives (pp. 7297). New York: Guilford Press.
Tinsley, H. E. A., & Tinsley, D. J. (1987). Uses of factor analysis
in counseling psychology research. Journal of Counseling Psy-

chology, 34, 414-424.

Ward, J. H. (1963). Hierarchical grouping to optimize an objective
function. Journal of the American Statistical Association, 58,
West, M., & Sheldon, A. E. (1988). Classification of pathological
attachment patterns in adults. Journal of Personality Disorders,

2, 153-159.

Received July 5, 1994

Revision received December 15, 1994
Accepted December 15, 1994