What are the main points/concepts of attachment theory as described in the article
Journal of Personality and Social Psychology 1991. “WJI. 61, No. 2, 226-244
Copyright 1991 by the American Psychological Association, Inc. 0022-3514/91/S3.00
Attachment Styles Among \bung Adults: A Test of a Four-Category Model
Save your time - order a paper!
Get your paper written from scratch within the tight deadline. Our service is a reliable solution to all your troubles. Place an order on any task and we will take care of it. You won’t have to worry about the quality and deadlines
Order Paper NowKim Bartholomew Simon Eraser University
Burnaby, British Columbia, Canada
Leonard M. Horowitz Stanford University
A new 4-group model of attachment styles in adulthood is proposed. Four prototypic attachment patterns are defined using combinations of a person’s self-image (positive or negative) and image of others (positive or negative). In Study 1, an interview was developed to yield continuous and categori- cal ratings of the 4 attachment styles. Intercorrelations of the attachment ratings were consistent with the proposed model. Attachment ratings were validated by self-report measures of self-con- cept and interpersonal functioning. Each style was associated with a distinct profile of interper- sonal problems, according to both self- and friend-reports. In Study 2, attachment styles within the family of origin and with peers were assessed independently. Results of Study I were replicated. The proposed model was shown to be applicable to representations of family relations; Ss’ attachment styles with peers were correlated with family attachment ratings.
This article describes a new model of attachment styles in adulthood. Drawing on the theory of Bowlby (1973, 1980, 1982a), two types of internal working models are postulated— an internal model of the self and an internal model of others. Each internal model can be dichotomized as positive or nega- tive to yield four theoretical attachment styles. This article sum- marizes the relevant childhood attachment literature, reviews recent work on adult attachment, describes the new model, and then presents two empirical studies designed to validate the proposed model.
Chi ldhood Attachment and Internal Models
Attachment theory conceptualizes “the propensity of human beings to make strong affectional bonds to particular others” (Bowlby, 1977, p. 201). Bowlby hypothesizes that an attachment system evolved to maintain proximity between infants and their caretakers under conditions of danger or threat. More recent formulations view the attachment system as functioning continuously to provide children with a sense of “felt security” which facilitates exploration by the child (Ainsworth, Blehar, Waters, & Wall, 1978; Sroufe & Waters, 1977). The quality of early attachment relationships is thus rooted in the degree to which the infant has come to rely on the attachment figure as a source of security (Ainsworth et al., 1978).
On the basis of infants’ responses to separation from and
This article is based on Kim Bartholomew’s doctoral dissertation at Stanford University
Preparation of this article was supported by Social Sciences and Humanities Research Council of Canada doctoral and postdoctoral fellowships to Kim Bartholomew.
We thank James Fuendeling for helping to develop and conduct in- terviews and Diana Doumas, Christine Duffy, Mike Emerzian, Gail Halloway, Stephanie Kondik, and Suzanne Little for coding inter- views.
Correspondence concerning this article should be addressed to Kim Bartholomew, Department of Psychology, Simon Fraser University, Burnaby, British Columbia, Canada V5A 1S6.
reunion with caretakers in a structured laboratory procedure, Ainsworth identified three distinct patterns of infant attach- ment: secure, anxious-resistant, and avoidant. Children classi- fied as securely attached welcome their caretaker’s return after a separation and, if distressed, seek proximity and are readily comforted. Infants classified as anxious-resistant show ambiva- lent behavior toward caregivers and an inability to be com- forted on reunion. Infants classified as avoidant avoid proxim- ity or interaction with the caretaker on reunion. Continuity in infant attachment patterns seems to be mediated largely by continuity in the quality of primary attachment relationships (see Lamb, Thompson, Gardner, Charnov, & Estes, 1985).
According to Bowlby’s theory, children, over time, internalize experiences with caretakers in such a way that early attachment relations come to form a prototype for later relationships out- side the family. Bowlby (1973) identifies two key features of these internal representations or working models of attachment: “(a) whether or not the attachment figure is judged to be the sort of person who in general responds to calls for support and protection; [and] (b) whether or not the self is judged to be the sort of person towards whom\anyone, and the attachment fig- ure in particular, is likely to respond in a helpful way” (p. 204). The first concerns the child’s image of other people; the second concerns the child’s image of the self. Recent research has exam- ined the nature of internal working models in relation to chil- dren’s earlier attachment styles. The data show, for example, that children classified as ambivalent hold negative views of themselves, but the data are not as consistent with respect to children classified as avoidant (Cassidy, 1988; Kaplan & Main, 1985; Main, Kaplan, & Cassidy, 1985). A considerable body of research also links the child’s attachment style at 12 or 18 months to the child’s social and emotional adjustment through early childhood (see Bretherton, 1985).
Attachment in Adulthood
A basic principle of attachment theory is that attachment relationships continue to be important throughout the life span (Ainsworth, 1982,1989; Bowlby, 1977,1980,1982b). Although
226
ATTACHMENT STYLES 227
evidence exists documenting the continuity of attachment-re- lated behaviors (see Belsky & Pensky, 1988; Bowlby, 1973,1980; Ricks, 1985; Rutter, 1988), investigators have only recently ex- amined the relationship between working models of attach- ment and social and emotional adaptation in adults. Main has developed an Adult Attachment Interview (George, Kaplan, & Main, 1987; Main & Goldwyn, 1988) that explores adults* repre- sentations of childhood attachment relations. On the basis of these interviews, mothers have been classified into attachment groups that parallel the three childhood attachment patterns described above and are predictive of the quality of the mother’s interaction with her own child and the security of the child’s attachment (Crowell & Feldman, 1987; Grossmann, Fremmer- Bombik, Rudolf, & Grossmann, 1988; Main et al, 1985). Ko- bak and Sceery (1988) used this procedure to examine young adults’ self- and other-representations, providing some evidence that secure subjects view themselves as relatively undistressed and others as supportive, that dismissive (avoidant) subjects view the self as undistressed and others as unsupportive, and that preoccupied subjects (corresponding to anxious-resistant children) view the self as distressed and others as supportive. (They did not discuss a category of people who might exhibit a negative view of both the self and others.)
Hazan and Shaver (1987) conceptualized romantic love as an attachment process and developed a self-report procedure to classify adults into three categories that correspond to the three attachment styles of childhood. In contrast to Main’s proce- dure, these investigators relied on respondents’ self-reports rather than on inferences from a semi-structured interview. Their results showed that compared with the secure group, the two insecure groups reported more negative experiences and beliefs about love, had a history of shorter romantic relation- ships, and provided less favorable descriptions of their child- hood relationships with parents (see also Collins & Read, 1990). Subjects in the two insecure groups also reported more self- doubt and less acceptability to others than did those endorsing a secure self-description (see also Feeney & Noller, 1990).
These two approaches differed both in the particular attach- ment relationships focused on (parent-child versus love rela- tionships) and in the methodology used for classifying subjects (interview versus self-report). Whereas the interview method identified avoidant adults as people who denied experiencing subjective distress and downplayed the importance of attach- ment needs, the self-report method identified people who re- ported feeling subjective distress and discomfort when they be- come close to others. Thus, a single avoidant-detached category may obscure conceptually separable patterns of avoidance in adulthood. Moreover, although Bowlby (1973) suggested that working models differ in terms of images of self and others, no study has considered all four categories that are logically de- rived by combining the two levels of self-image (positive vs. negative) with the two levels of image of others (positive vs. negative). The present research examined all four of these cate- gories and assessed subjects through an interview as well as through subjects’ own self-reports.
A Model of Adult Attachment
The model of the self and the model of the other as conceptu- alized by Bowlby can be combined to describe prototy pic forms
of adult attachment (Bartholomew, 1990). If a person’s abstract image of the self is dichotomized as positive or negative (the self as worthy of love and support or not) and if the person’s ab- stracted image of the other is also dichotomized as positive or negative (other people are seen as trustworthy and available vs. unreliable and rejecting), then four combinations can be con- ceptualized. Figure 1 shows the four attachment patterns that are derived from a combination of the two dimensions. Each cell represents a theoretical ideal, or prototype (Cantor, Smith, French, & Mezzich, 1980; Horowitz, Wright, Lowenstein, & Parad, 1981; Rosen, 1978), that different people might approxi- mate to different degrees.
Cell I indicates a sense of worthiness (lovability) plus an ex- pectation that other people are generally accepting and respon- sive. Because this cell corresponds conceptually to categories that investigators call securely attached (e.g., Hazan & Shaver, 1987; Main et al, 1985), we labeled it secure. Cell II indicates a sense of unworthiness (unlovability) combined with a positive evaluation of others. This combination of characteristics would lead the person to strive for self-acceptance by gaining the ac- ceptance of valued others. This pattern corresponds concep- tually to Hazan and Shaver’s ambivalent group (Hazan & Shaver, 1987) and to Main’s enmeshed or preoccupied with at- tachment pattern (Main et al., 1985) and is referred to as preoc- cupied. Cell III indicates a sense of unworthiness (unlovability) combined with an expectation that others will be negatively disposed (untrustworthy and rejecting). By avoiding close in- volvement with others, this style enables people to protect themselves against anticipated rejection by others. Although not explicitly discussed in previous work in adult attachment, this style may correspond in part to the avoidant style described by Hazan and Shaver (1987). We therefore labeled it fearful- avoidant. Finally, Cell IV indicates a sense of love-worthiness combined with a negative disposition toward other people. Such people protect themselves against disappointment by avoiding close relationships and maintaining a sense of inde- pendence and invulnerability. This style corresponds concep- tually to the detached or dismissing of attachment attitude de- scribed by Main et al. (1985), so we labeled it dismissive- avoidant.
The dimensions in Figure 1 can also be conceptualized in
MODEL OF SELF (Dependence)
Positive (Low)
Negative (High)
Positive (Low)
MODELOF OTHER (Avoidance)
Negative (High)
CELL I
SECURE Comfortable with
intimacy and autonomy
CELL IV
DISMISSING Dismissing of intimacy
Counter-dependent
CELL II
PREOCCUPIED Preoccupied with
relationships
CELL HI
FEARFUL Fearful of intimacy Socially avoidanl
Figure 1. Model of adult attachment.
228 KIM BARTHOLOMEW AND LEONARD M. HOROWITZ
terms of dependency on the horizontal axis and the avoidance of intimacy on the vertical axis (see labels in parentheses). De- pendency can vary from low (a positive self-regard is estab- lished internally and does not require external validation) to high (positive self-regard can only be maintained by others’ on- going acceptance). Avoidance of intimacy reflects the degree to which people avoid close contact with others as a result of their expectations of aversive consequences. The dismissing and fearful styles are alike in that both reflect the avoidance of intimacy; they differ, however, in the person’s need for others’ acceptance to maintain a positive self-regard. Similarly, the preoccupied and fearful groups are alike in that both exhibit strong dependency on others to maintain a positive self-regard, but they differ in their readiness to become involved in close relationships. Whereas the preoccupied cell implies a reaching out to others in an attempt to fulfill dependency needs, the fearful cell implies an avoidance of closeness to minimize even- tual disappointment. Therefore, cells in adjoining quadrants of Figure 1 are more similar conceptually than those in opposite quadrants.
Study 1
We administered a semi-structured interview asking subjects to describe their friendship patterns; the subjects’ responses were used to assess the degree to which each person approxi- mated each of the four styles in Figure 1. We also obtained self-report and friends’ ratings of each attachment style, so the three sources of data could be compared. Additional question- naires were administered to subjects and their friends to test hypotheses implied by the model. In particular, self-reports of self-concept, sociability, and interpersonal problems were ob- tained, and self-reports of interpersonal problems were corrob- orated by the judgments of close friends.
The study tested three specific hypotheses. First, multidi- mensional scalings of each set of ratings (by raters, subjects, and friends) were expected to reproduce the organization of Figure 1 across all three sources of data (interview, self-report, and friend-report). Second, self-concept measures were expected to differentiate groups with a positive model of the self (secure and dismissing) from those with a negative model of the self (preoc- cupied and fearful), whereas a sociability measure was expected to differentiate groups with a positive model of others (secure and preoccupied) from those with a negative model of others (fearful and dismissing). Third, the groups were expected to differ from each other in their interpersonal problems. The groups with a negative image of self (preoccupied and fearful) were expected to exhibit problems with passivity and unasserti- veness, whereas those with a negative image of others (fearful and dismissing) were expected to describe problems with socia- lizing and intimacy Problems described by the secure group were not expected to be distinctive in content,
Method
Subjects
Forty female and 37 male students from an introductory psychology class constituted the target sample. They ranged in age from 18 to 22 (M = 19.6); 67% were White, 16% Asian, 5% Hispanic, 8% Black, and
4% other. An equal number of same-sex friends constituted the friend sample. The friends’ age ranged from 18 to 23 (M = 19.8); 65% were White, 13% Asian, 13% Hispanic, 4% Black, and 5% other.
Procedure
Equal numbers of men and women were randomly selected from the subject pool and contacted by telephone. First-year undergraduate stu- dents were excluded to ensure that subjects had been at college long enough to make close friends. Potential subjects were invited to partici- pate with a friend in “a study of friendship and how well people know each other.” Friends were required to be close, same-sex, nonromantic friends whom subjects had known for at least 6 months. Subjects re- ceived course credit for participation, and their accompanying friends were paid $5.
Subjects were tested in groups of two to four friendship pairs. Sub- jects and their friends completed two sets of questionnaires, one re- questing demographic and personality information about themselves and the other asking the same questions of their partner. Instructions were given for participants to answer the second set of questions “ac- cording to your perceptions and knowledge of your friend’s character, feelings or behavior, and not according to how you think your friend may be likely to answer them.” Then each subject and friend were separated to complete the questionnaires. To avoid contrast effects, target subjects completed the self-report questionnaires first, and their friend completed the friend-report first. A second session was then scheduled for subjects to receive the interview.
Measures
Attachment interview. The first author administered a semi-struc- tured interview, which lasted approximately 60 min, to subjects in the target sample. Each interview was tape recorded. The interviewer asked subjects to describe their friendships, romantic relationships, and feelings about the importance of close relationships. If subjects had not been involved in a romantic relationship, they were asked the rea- sons. They were asked about loneliness, shyness, their degreeof trust of others, their impressions of other people’s evaluations of themselves, and their hopes for any changes in their social lives.
On the basis of the interview audio recordings, three raters indepen- dently rated each subject on four 9-point scales describing the subject’s degree of correspondence with each of the four prototypes. The raters comprised two advanced female undergraduate psychology majors and one female graduate student; they were blind to all other measures in the study. A set of criteria (available from the first author) described each prototype, and the rater was instructed to judge how well a sub- ject’s responses matched each of the prototypic descriptions. The four prototypes can be briefly summarized as follows. The secure prototype is characterized by a valuing of intimate friendships, the capacity to maintain close relationships without losing personal autonomy, and a coherence and thoughtfulness in discussing relationships and related issues. The dismissing prototype is characterized by a downplaying of the importance of close relationships, restricted emotionality, an em- phasis on independence and self-reliance, and a lack of clarity or credi- bility in discussing relationships. The preoccupiedprototype is charac- terized by an overinvolvement in close relationships, a dependence on other people’s acceptance fora sense of personal well-being, a tendency to idealize other people, and incoherence and exaggerated emotiona- lity in discussing relationships. The fearful prototype is characterized by an avoidance of close relationships because of a fear of rejection, a sense of personal insecurity, and a distrust of others. Alpha coefficients were computed to assess the reliability of the prototype ratings. The reliabilities ranged from .87 to .95. The ratings were averaged, and the highest of the four average ratings was considered to be the best-fitting
ATTACHMENT STYLES 229
category for that subject. From this procedure, 47% of the sample was classified as secure, 18% as dismissing, 14% as preoccupied, and 21 % as fearful. In addition, the raters were asked to rate each interview along 15 dimensions of relevance to adult attachment (see Appendix A for definitions of the dimensions).
Self- and friend-reports. All subjects completed demographics and friendship questionnaires, two self-concept measures, and a sociability measure. In addition, the subjects completed the Relationship Ques- tionnaire and the Inventory of Interpersonal Problems twice, once to describe themselves and once to describe their friend. The friend’s version of these questionnaires was identical to the standard self-re- port version, with the exception of wording: Instead of “I try to please other people too much,” for example, the friend’s version read “F[your friend ] tries to please other people too much.” The following question- naires were used:
1. The Demographics Questionnaire included family information (e.g., marital status of parents, number of siblings) and personal activi- ties (e.g., exercise, religious observance). Seven-point items assessed the degree of experienced depression, anxiety, and happiness (with re- versed scoring). These three items were combined into a composite measure of subjective distress ifx = .68).
2. The Friendship Questionnaire contained equivalent demographic, factual, and personal questions about the friend. One item assessed the duration of the friendship, and five items assessed the nature of the friendship (e.g., “Compared with close friendships you’ve had in the past, how close is your friendship with FT*). The latter five items were combined into a friendship closeness scale (target sample a = .80; friend sample a = .86).
3. The Rosenberg Self-Esteem Inventory (Rosenberg, 1965) is a 10- item scale that measures global self-esteem. A sample item is “I cer- tainly feel useless at times” (reverse scored; coefficient a = .85).
4. The Fey Self-Acceptance Scale(Fey, 1955) is a 20-item measure of self-acceptance. A representative item is “I’m pretty satisfied with the way I am” (coefficient a = .86).
5. The Sociability Scale (Cheek & Buss, 1981) is a 5-item measure that assesses the degree to which people like to socialize with others. A sample item is “I like to be with people” (self-report coefficient a = .74; friend-report coefficient = .78).
6. The Relationship Questionnaire is an adaptation of the attach- ment measure developed by Hazan and Shaver (1987). This measure consists of four short paragraphs describing the four attachment styles (see Appendix B). Each respondent is asked to make ratings on a 7- point scale of the degree to which they (or their friend) resemble each of the four styles. These ratings are referred to as the self-report and friend-report attachment ratings.
7. The Inventory of Interpersonal Problems (IIP; Horowitz, Rosen- berg, Baer, Ureno, & Vitlasenor, 1988; Horowitz, Rosenberg, Ureno, Kalehzan, & O’Halloran, 1989) is a 127-item inventory designed to assess interpersonal difficulties in a broad cross-section of interper- sonal domains. Subjects are asked to describe the amount of distress that they have experienced from each interpersonal problem on a 5- point scale ranging from not at all (0) to extremely (4). The IIP has demonstrated acceptable reliability, validity, and sensitivity to changes that occur during psychotherapy (Horowitz et al., 1988; 1989). Re- sponses on this inventory were scored using a circumplex. procedure that locates subjects’ responses within a two-dimensional interper- sonal space (defined by the dimensions of warmth and dominance) that can be divided into eight octants (Kiesler, 1983; Wiggins, 1979). Sepa- rate subscales describe distress in each of these octants (Alden, Wig- gins, & Pincus, 1990). Starting with the region of greatest dominance (neutral on warmth) and moving counterclockwise, the eight octants are labeled overly autocratic (e.g., “I try to control other people too much”), overly competitive (e.g., “I fight with other people too much”), overly cold (e.g., “I keep other people at a distance too much”), overly
introverted (e.g., “I feel embarrassed in front of other people too much”), overly subassertive (e.g., “It is hard for me to be assertive with another person”), overly exploitable (e.g., “I let other people take advantage of me too much”), overly nurturant (e.g., “I put other people’s needs before my own too much”), and overly expressive (e.g., “I want to be noticed too much”). (Self-report subscales coefficient a = .66 to .88; friend-report subscales = .71 to .89.)
Results
Description of Attachment Groups
The 15 dimensions rated in the attachment interview were used to identify a profile of correlates for each of the four at- tachment prototypes. Because the prototype ratings were based, in part, on subjects’ dimension scores, these data should be regarded as primarily descriptive. Table 1 shows mean rat- ings on each of the 15 dimensions rated in the attachment inter- view for subjects classified according to their highest prototype rating, as well as the results of one-way analyses of variance and Newman-Keuls comparisons for each dimension. Significant group differences existed for each of the 15 dimensions. Corre- lations between each dimension and each of the four continu- ous prototype ratings were also calculated.
As shown in Table 1, the secure group obtained uniquely high ratings on the coherence of their interviews and the degree of intimacy of their friendships. They also received high ratings on warmth, balance of control in friendships, and level of involve- ment in romantic relationships. Correspondingly; the secure prototype correlated highly with ratings of coherence, r(75) = .78, p < .001, intimacy, r(75) = .77, p < .001, balance of control in friendships, r(15) = .29, p < .01, level of involvement in romantic relationships, r{15) = .40, p < .001, self-confidence, r(75) = .41, p < .001, and warmth, r(75) = .59, p < .001.
The dismissing group scored uniquely high on self-confi- dence and uniquely low on emotional expressiveness, frequency of crying, and warmth. They scored lower than the secure and preoccupied on all scales reflecting closeness in personal rela- tionships: self-disclosure, intimacy, level of romantic involve- ments, capacity to rely on others, and use of others as a secure base. They were also rated as being low on elaboration and caregiving and as being more in control than their social partners in both friendships and romantic relationships. Con- tinuous ratings of the dismissing prototype correspondingly correlated positively with self-confidence, r(75) = .52, p < .001, and negatively with elaboration, r{15) = – .43, p < .001, emo- tional expressiveness, r{15) = -.69, p < .001, frequency of cry- ing, r(75) = -.55, p < .001, warmth, r(75) = -.68, p < .001, caregiving, r(75) = -.37, p < .001, and with all scales signifying involvement in close relationships—self-disclosure, r{75) = -.45, p < .001, intimacy, r(15) = -.33, p < .01, level of romantic involvements, r{15) = -.35, p < .001, reliance on others, r(15) = -.57, p < .001, and use of others as a secure base, r(75) = -.57, p < .001. Consistent with the group results, the dismissing rat- ing was also positively associated with the balance of control scales, K75) = .40, p < .001, and r(75) = .39, p < .001 for friend- ships and for romantic relationships, respectively.
The profile shown by the preoccupied group was opposite to that of the dismissing in almost every respect; the means of the two groups differed significantly on each of the 15 rating
230 KIM BARTHOLOMEW AND LEONARD M. HOROWITZ
Table 1 Mean Interview Ratings Across Attachment Groups
Model
Self Other
n
General Elaboration Coherence
Friendships Self-disclosure Intimacy Balance of control
Romantic relationships Highest level involvement Balance of control
Personal characteristics Self-confidence Emotional expressiveness Crying frequency
Interpersonal characteristics Warmth Reliance on others Others as secure base Nonsocial vs social crying Caregiving
Secure
Positive Positive
36
6.84, 6.08c
4.00b 5.09c 2.94C
3.6 le 3.29b
3.54C 3.76b 2.52b
3.13C 4.25b 3.34b 2.50b 4.25b
Dismissing
Positive Negative
14
Measure
5.74, 5.45b
3.33, 4.10, 3.10e
2.60B 3.73C
3.93d 2.60. 1.45.
2.21, 3.268 2.43a 2.17,, 3.79,
Preoccupied
Negative Positive
11
7.95C 4.76,
5.12C 4.54b 2.62fl
3.61C 2.91a,b
2.86b 5.2Oc 3.32C
2.94^ 5.I5C 4.I2C 3.70c 4.80c
Fearful
Negative Negative
16
6.23^ 5.00^b
3.05, 3-79, 2.608
2.60, 2.72a
2.24a 3.39b 2.17,,
2.70,, 3.46, 2.61. l.75b 4.25^,
I\% 76)
6.90* 16.54*
19.40* 23.12*
7.94*
6.06* 8.50*
27.77* 12.32* 5.75*
11.77* 14.65* 21.09*
7.60* 6.51*
Note, Means with different subscripts differ significantly at p < .05 according to Newman-Keuls pairwise comparisons. *p<.001.
scales. The preoccupied group scored uniquely high on ela- boration, self-disclosure (showing a tendency to disclose inap- propriately), emotional expressiveness, frequency of crying, reli- ance on others, use of others as a secure base, crying in the presence of others, and caregiving. They were also rated high on level of romantic involvement and low on coherence and bal- ance of control in friendships. Continuous ratings of the preoc- cupied prototype correlated positively with the presence of elab- oration, r(75) = .61, p < .001, emotional expressiveness, r(75) = .78, p < .001, level of romantic involvements, r(75) = .34, p < .01, disclosure, ii75) = .60, p < .001, the tendency to rely on others, r(75) = .64, p < .001, use others as a secure base, r(15) – .60, p < .001, and caregiving, r{75) = .38, /? < .001. The preoccu- pied rating was also positively correlated with the tendency to cry frequently, r{75) = .61, p < .001, and in the company of others, r(75) = .45, p < .001, and negatively correlated with the balance of control in friendships, r(75) = -.35, p < .01, coher- ence, r(75) = -.39, p < .001, and self-confidence, r{75) = – .33, p<.001.
Finally, the fearful group was rated significantly lower than the secure and preoccupied on self-disclosure, intimacy, level of romantic involvement, reliance on others, and use of others as a secure base when upset. They were also rated as uniquely low in self-confidence and as low on both balance-of-control scales. Continuous ratings of the fearful prototype showed negative correlations with self-confidence, r(75) – -.70, p < .001, and
coherence, r{75) ~ -.35, p < .01, and with all measures indicat- ing closeness of relationships—including self-disclosure, K75) = -.43, p < .001, intimacy, r(75) = -.52, p < .001, level of involvement in romantic relationships, r(75) = —.36, p < .01, capacity to rely on others, r(75) = -.30, p < .01, and use of others as a secure base, r(75) = -.33, p < .01. The fearful rating was also negatively correlated with the balance of control mea- sures for both friends, r(75) = —.40, p < .001, and romantic relationships, r(75) = —.32, p < .01, indicating a tendency to assume a subservient role in close relationships.
A discriminant analysis was performed to assess the degree to which the various interview ratings accounted for the overall discrimination between the four attachment groups. The analy- sis (using a stepwise method with minimization of Wilks’s lambda as the selection criterion) resulted in three significant discriminant functions, which correctly classified 92% of the sample—including 86% of the secure group, 94% of the fearful group, and 100% of both the preoccupied and dismissing groups. Twelve of the 15 interview rating scales independently contributed significant variance to the final functions, con- firming that the group distinctions are multiply determined.
Sex Differences in Attachment Ratings
Female subjects received significantly higher ratings than male subjects on the interview-based preoccupied rating
ATTACHMENT STYLES 231
(women’s M= 3.10, men’s M = 2.00), z(75) = 2.88, p < .01; male subjects received significantly higher ratings than female sub- jects on the interview-based dismissing rating (women’s M = 3.10, men’s M= 4.01), t(15) = 2.70, p < .01. To control for these sex differences, sex of subject was included as a covariate in all correlational and group analyses.
Depth of Friendships
On average, the pairs of friends reported reasonably close friendships of 1 to 2 years’ duration. The four groups did not differ significantly in the length of the friendships according to either subjects, F(3, 72) = 1.22, ns, or their accompanying friend, F(3, 72) = 0.75, ns. In addition, there were no group differences in the composite measure of closeness of friend- ships according to either subjects, F(3,72) = 1.82, ns, or friends, F(3, 72) = 0.37, ns. Therefore, the group differences reported below cannot be explained by a difference in either the length or the closeness of the friendships.
Intercorrelations of Attachment Styles
The pattern of correlations among the four attachment rat- ings for each method was hypothesized to be consistent with the model presented in Figure 1. Interview attachment ratings in opposing positions were negatively correlated: Between the secure and fearful ratings, r(75) = — .55, p< .001; between the preoccupied and dismissing ratings, r(15) = -.50, p < .001. The correlations between styles in adjacent positions showed non- significant or low negative correlations with one another (range = —.26 to —.14). Similar patterns were found for the self-report and friend-report attachment ratings (on the Rela- tionship Questionnaire): The secure and fearful ratings were negatively correlated (respective rs = -.65 and -.69, ps < .001), and the preoccupied and dismissing ratings were negatively correlated (respective rs = -.37 and —.41, ps < .001). For both self- and friend-reports, the styles in adjacent positions showed nonsignificant or negative correlations with one another with one exception: Friend ratings of the fearful and dismissing styles were positively correlated, r(J5) = .27, p < .05.
Multidimensional scalings were performed to determine if the dimensional structure underlying the intercorrelations of the attachment ratings were consistent with the proposed model. Intercorrelations among the four continuous attach- ment ratings were used as measures of proximity. A separate correlation matrix was prepared for each set of attachment rat- ings—interview ratings, the subjects’ own ratings of their match to each attachment style (on the Relationship Question- naire), and the friends’ ratings of subjects on each attachment style (on the friend version of the Relationship Questionnaire). Each matrix was then subjected to a nonmetric multidimen- sional analysis using the program KYST (Kruskal, Young, & Seery, 1973). The one-dimensional solutions yielded stress val- ues from .38 to .50 (Stress Formula 2; Kruskal & Wish, 1978), whereas the two-dimensional solutions yielded a stress value of .00 in each case. To facilitate comparisons with the theoretical model, the axes were rotated so that the secure placement was in the upper left quadrant at 45° from each axis. As indicated in
Figure 2, each set of data yielded a configuration roughly corre- sponding to the structure proposed in Figure 1.
In addition, a factor analysis was performed to examine the convergence of interview ratings, the subjects* self-ratings, and the friends’ ratings. Figure 3 shows the results of a principal- components analysis with varimax rotation of the intercorrela- tions of the three sets of attachment ratings (with axes rotated to facilitate interpretation). The two factors accounted for 47% of the variance.1
Self-Concept Measures
Subjects in the two groups that were theoretically expected to reflect a negative self-image were hypothesized to exhibit lower scores on measures of self-concept than subjects in the two groups that were theoretically expected to reflect a positive self- image. Using subject classifications derived from the interview- based attachment ratings, a 2 (positive vs. negative self-image) X 2 (positive vs. negative other-image) multivariate analysis of co- variance (MANCOVA; using sex as a covariate) was therefore performed on the three self-concept measures: self-esteem, self- acceptance, and subjective distress. The results showed a signifi- cant main effect for the self factor only, as hypothesized, FQ, 70) = 7.11, p < .001. Follow-up univariate analyses showed significant effects for the self factor for each of the three mea- sures (all ps < .001; see Table 2 for group means). No other main effect or interaction was significant.
Correlational analyses confirmed this pattern of results. Rat- ings of the degree to which subjects matched the secure and dismissing prototypes were positively correlated with measures of self-concept (partial correlations, controlling for sex, ranged from .20 to .41, all ps < .05), whereas ratings of the degree to which subjects matched the fearful and preoccupied prototypes were negatively correlated with measures of self-concept (par- tial correlations ranged from —.18 to -.49, all ps < .06).
Sociability Measure
On the other hand, the model implies that subjects in the two groups that were expected to reflect a positive image of others would exhibit higher scores on the measure of sociability than subjects in the two groups expected to reflect a negative image of others. Therefore, a 2 (positive vs. negative self-image) X 2 (positive vs. negative other-image) analysis of covariance (AN- COVA) was performed on the sociability measure (with sex as a covariate). The results showed a significant main effect for the other factor only, as hypothesized, F(l, 72)= 19.96,p<.001 (see Table 2 for group means). Correspondingly, sociability was posi- tively correlated (controlling for sex) with the secure and preoc- cupied prototype ratings, r(14) = .36, p < .001, and r{74) = .24, p < .05, and negatively correlated with the fearful and dismiss- ing prototype ratings, r{lA) = – .41, p < .001, and r(74) = -.20, p < .05.
1 Although the different sources of attachment ratings tended to con- verge, the correlations between specific attachment ratings across methods were not high. The average correlation between correspond- ing ratings across interview and self-reports was. 34; the average corre- lation across interview and friend-reports was .25.
232 KIM BARTHOLOMEW AND LEONARD M. HOROWITZ
Interview
\X>*
0 . 3 –
.1.0 •
Self-report Friend-report
1.0-
0 J –
•OS-
•
•in-
secure
A
A
Dismissing
Preoccupied
A
A
Fearful
1.0-
0.3-
0 –
-0.3-
-1.0 •
Secure
A
A Dismissing
L – ~ i 1 1 1
Preoccupied
A
A
Fearful ‘ ” 1 ” – ‘ 1 -T” ‘ 1 ”
• iJO -OS 0 0.3 1 4 ‘ * -0-3 0
Figure 2. Multidimensional scalings of interview, self-report, and friend-report attachment ratings.
0.5 1.0
ATTACHMENT STYLES 233
Secure ^
r Is
\ 1 1 -0.8
Dismissing
.-—”
-0.4
/ a /
-0.4
-0.8
0.8
0.4
1
>
L
/
0.4
>
Preoccupied
^ aj
i i i i i i
0.8
y/ Fearful
O Interview ratings
A Self-report ratings
• Friend-report ratings
Figure 3. Factor analysis of interv iew, self-report, and friend-report attachment ratings.
Self-report ratings of sociability were corroborated by friend- reports. The main effect for positivity of other models was repli- cated, F(\, 72) = 7.08, p < .001, but there was also a significant effect for the self model factor, F{\, 72) = 3.12, p < .05. Inspec- tion of means (see Table 2) indicated that friends tended to see the dismissing group as especially low on sociability: Follow-up Newman-Keuls comparisons indicated that the mean of the dismissing group was significantly lower than that of each of the other attachment groups.
Interpersonal Problems
Finally, we examined the location within the two-dimen- sional interpersonal space of subjects’ interpersonal problems, as reported on the UP. First, each subject’s mean response across all problems was examined. A two-way ANCOVA re- vealed a significant main effect for the self factor only, F(l, 72) = 11.82, p < .001. The two groups with a negative self-image re- ported higher levels of interpersonal problems overall than the groups with a positive self-image. Parallel analyses with friends’ reports revealed a trend for groups with negative self models to be judged to experience higher mean levels of interpersonal problems, F(l, 71) = 2.93, p < .10. Therefore, in the following analyses, each subject’s score on each of the eight subscales (for both self- and friend-reports) was converted to an “ipsative score.” That is, each subject’s subscale scores were expressed as a
deviation from that subject’s overall mean, thereby reflecting the extent to which that group of problems was more or less problematic for that particular person.2 Each subscale score was then expressed as a z score using the mean and standard deviation of that subscale for the entire sample.
Each of the eight subscale scores was correlated with each prototype rating, and the mean of each subscale was also com- puted for each prototype group. Group means on the circum- plex subscales, for self- and friend-reports, are shown graphi- cally in Figure 4. Significant group differences were found on six of the eight subscales for both self-reports and friend-re- ports. Correlational analyses for self-reports and friend-reports are contained in Table 3. All of the self-report and friend-report problem subscales were significantly correlated with at least one of the attachment ratings.
As shown in Figure 4, the secure group’s profile of means was elevated on the warm side of the interpersonal space according to both self- and friend-reports. However, no one subscale score was extreme: The largest standardized mean scores were on the order of 0.25. Correlations between secure ratings and the self- report circumplex subscales revealed significant, although mod- est, positive correlations with the overly expressive and autocra- tic scales, as well as negative correlations with the cold and introverted scales. The friends’ reports showed positive correla- tions with the exploitable and nurturant scales and negative correlations with the cold and introverted scales.
The dismissing group showed a self-reported profile centered on the hostile side of the interpersonal space. Correlational analyses indicated that the cold subscale is most highly corre- lated with this attachment style. The friends generally con- firmed this pattern, although friends1 reports tended to show the dismissing style as more strongly associated with introver- sion than self-reports did. Also, the dismissing rating was nega- tively correlated with both self- and friend-ratings of the nurtur- ant and expressive scales and with self-ratings of the subasser- tive and exploitable scales.
The self-report profile of the preoccupied group showed an elevation on the overly expressive scale, with the correlational analyses confirming this location of interpersonal problems. Friends’ reports also described these subjects as highest on the
2 When a personality inventory has many subscales, it is sometimes necessary to ipsatize subjects’ subscale scores. This procedure has been shown to be particularly important for the IIP because a principal- components analysis of the IIP items typically yields a large general first factor that reflects differences among people in their readiness to endorse complaints (Horowitz, Rosenberg, Baer, Ureno, & Villasenor, 1988). That factor may be considered a complaint factor, an acquies- cence factor, or an intensity factor, and as shown by Wiggins, Steiger, and Gaelick (1981, p. 283), it reflects individual differences in the use of the response format, rather than differences in important aspects of interpersonal functioning. Therefore, the person’s average response level must be treated separately from the circumplex components per se. The preferred way to control for individual differences in people’s average response level is to ipsatize the subscale scores by expressing each score as a deviation from the subject’s own mean across all sub- scales (Cronbach, 1949; Horowitz et al., 1988; Strack, 1987; Wiggins & Pincus, 1989). The ipsatizing procedure has been shown to improve the circumplex properties of interpersonal measures (Paddock & No- wicki, 1986; Rinn, 1965; Wiggins et al., 1981).
234 KIM BARTHOLOMEW AND LEONARD M. HOROWITZ
Table 2 Means of Self-Report Self-Concept Measures and Self and Friend-Report Sociability Across Attachment Groups
Model
Self Other
n
Self-concept Distress Self-esteem Self-acceptance
Sociability Self-report Friend-report
Secure
Positive Positive
36
3.34 35.28
3.75
3.62 3.48
Dismissing
Positive Negative
14
Measure
3.00 36.50
3.87
3.07 2.63
Preoccupied
Negative Positive
11
4.30 32.21
3.39
3.87 3.71
Fearful
Negative Negative
16
4.19 30.69
3.19
2.91 3.31
Note. Higher numbers reflect higher distress, self-esteem, self-acceptance, and sociability.
overly expressive scale, but the friends’ means were also ele- vated in other octants reflecting dominance (i.e.; the autocratic and competitive scales). Correspondingly, the preoccupied rat- ing was positively associated with friend-ratings of the expres- sive, autocratic, and competitive subscales. That is, the friends’ descriptions of the subjects’ problems revealed less overall warmth and more overall dominance than the subjects1 self-re- ports revealed. Preoccupied ratings were also negatively corre- lated with self-ratings on the cold scale and friend-ratings on the introverted and exploitable scales.
Finally, the fearful group reported relatively more problems reflecting a lack of assertiveness and social inhibition (introver- sion). The corresponding friends’ reports confirmed this gen- eral pattern of results, as did the correlational analyses for both sets of data. Also, the fearful rating was negatively correlated with both self- and friend-ratings of the autocratic scale and with self-ratings of the competitive and expressive scales.
Conclusions
This study demonstrated that a semi-structured interview can be used to assess the degree to which subjects approximate each of the four hypothesized attachment styles. An additional 15 interview rating scales demonstrated the distinctiveness of each of the four styles. The internal structures of the interview ratings, the self-report ratings, and the friend-report ratings of the four attachment styles were consistent with the proposed model. Thus, a two-dimensional structure spatially reproduced the hypothesized relationships among the four styles, and the three sets of ratings independently produced similar results. The data also supported the hypothesis that the two groups theoretically described as having a positive self-model would differ on measures of self-concept from the two groups theoreti- cally described as having a negative self-model. Also as hypoth- esized, the two groups theoretically described as having a posi- tive model of others differed on a measure of sociability from the two groups described as having a negative model of others.
Taken together, the results supported the implication of Bowlby’s theory that four different attachment styles can be identified.
The circumplex analyses confirmed the hypothesis that each style is associated with a distinct pattern of interpersonal prob- lems. Fearful subjects were most likely to report interpersonal problems in the overly passive region of the interpersonal space (the lower quadrants), whereas dismissing subjects were more likely to report problems related to a lack of warmth in social interactions. The one group whose problems did not conform to the original hypothesis was the preoccupied group. Whereas their negative self-image and positive other-image were ex- pected to reflect problems in being overly warm and passive (the lower right quadrant), their problems reflected a greater degree of warmth-dominance (the upper right quadrant). Ex- amples of items falling in this region of the interpersonal space are “I want to be noticed too much” and “it is hard for me to stay out of other people’s business.” These findings suggest that al- though preoccupied people are highly dependent on others to maintain positive self-regard, they attempt to achieve this aim though a controlling (overly dominating) interpersonal style. Finally, the data also showed that self- and friend-reports of interpersonal problems were generally consistent across the four attachment styles.
Study 2
Study 1 examined ratings of the attachment styles that were based on the subjects’ descriptions of their close friendships and romantic relationships, and these assessments were shown to predict important aspects of subjects’ personalities and inter- personal functioning. Study 2 was designed (a) to replicate the circumplex analyses of Study I, (b) to extend the proposed model of attachment to representations of relationships within the family of origin, and (c) to investigate the relationships be- tween family and peer attachment representations. To examine these questions, the subjects of Study 2 were interviewed about
ATTACHMENT STYLES 235 Autocratic Autocratic
Competitive
Introverted
Subassertive
Expressive
Exploitable
Competitive
Cold
Introverted
Expressive
Nurturant
Exploitable
Subassertive
Competitive
Introverted
DISMISSING
Autocratic
Expressive
Nurturant
Exploitable
Competitive
Introverted
FEARFUL
Autocratic
Expressive
Nurturant
Exploitable
Subassertive Subassertive