Self-Report Measurement of Adult Attachment: An Integrative Overview.
Self-Report Measurement of Adult Attachment: An Integrative Overview
Kelly A. Brennan
State University of New York at Stony Brook
Catherine L. Clark
Western Consortium for Public Health
Phillip R. Shaver
University of California, Davis
In J. A. Simpson & W. S. Rholes (1998) (Eds.), Attachment theory and close relationships (pp. 46-76). New York: Guilford Press. Address correspondence to the first author at the Department of Psychology, State University of New York at Brockport, Brockport, NY 14420. Electronic mail inquiries may be directed to
[email protected].
Self-Report Measurement of Adult Attachment: An Integrative Overview
Ever since Hazan and Shaver (1987) showed that it is possible to use a self-report questionnaire to measure adolescent and adult romantic-attachment orientations (secure, anxious, and avoidant--the three patterns identified by Ainsworth, Blehar, Waters, and Wall, 1978, in their studies of infant-caregiver attachment), a steady stream of variants and extensions of their questionnaire have been proposed. The resulting diversity often arouses frustration and confusion in newcomers to the field who wonder which of the many measures to use. The three of us are probably typical of attachment researchers in receiving as many as five telephone calls, letters, and e-mail messages a week from researchers who want to know either "Has anything happened since 1987?" or "Which measure is the best?" In the present chapter we attempt to solve this problem by creating an all-purpose reply to future attachment researchers who wish to use self-report measures. Interview measures have also been proposed, but we will say little about them here. Attachment interviews are powerful and perhaps uniquely revealing, but they are also impractical for most researchers. (See Bartholomew & Horowitz, 1991; Bartholomew & Shaver, this volume; Main, Kaplan, & Cassidy, 1985; Scharfe & Bartholomew, 1994; and van IJzendoorn, 1995, for discussions of attachment interview measures, not all of which measure the same constructs.)
Hazan and Shaver (1987, 1990) asked research participants to indicate which of three attachment-style prototypes (shown here in Table 1) best characterized their feelings and behavior in romantic relationships. These authors naively took for granted that Ainsworth et al. (1978) were correct in thinking of attachment patterns (usually called "attachment styles" by social psychologists) as categories or types. In retrospect, it is evident that Hazan and Shaver should have paid attention to Ainsworth et al.'s Figure 10 (p. 102), which summarized the results of a discriminant analysis predicting infant attachment type (secure, anxious, or avoidant) from the continuous rating scales used by coders to characterize the infants' behavior in a laboratory "Strange Situation." Our Figure 1 reproduces the essential features of the Ainsworth et al. figure and also includes our names for the two discriminant functions: Avoidance and Anxiety.
The coding scales that correlated most highly with the avoidance dimension (Function 1) were: (1) avoiding mother during episodes 5 and 8 of the Strange Situation (the two reunion episodes), (2) not maintaining contact with mother during episode 8, (3) not seeking proximity during episode 8, and (4) engaging in more exploratory behavior and more distance interaction (communication with a stranger while mother was absent) in episode 7 of the Strange Situation. All of these scales indicate avoidance of mother, lack of closeness to mother, and less distress during mother's absence (in the presence of an adult stranger). The coding scales that correlated most highly with the anxiety dimension (Function 2) were: (1) crying (all through episodes, 2-8, but especially episode 6, when the infant was left alone for 3 minutes), (2) greater angry resistance to mother during episodes 5 and 8 (the reunions), (3) greater angry resistance to the stranger during episodes 3, 4, and 7 (when the stranger tried to comfort or play with the infant), and (4) reduced exploration in episode 7, when the solitary infant was joined by a stranger.
----------------------------------------------
Insert Table 1 and Figure 1 about here
----------------------------------------------
Figure 1 indicates that, right from the start, Ainsworth's three major attachment "types" could be conceptualized as regions in a two-dimensional space, the dimensions being Avoidance (discomfort with closeness and dependency) and Anxiety (crying, failing to explore confidently in the absence of mother, and angry protest directed at mother during reunions after what was probably experienced as abandonment). When Levy and Davis (1988) first asked adult subjects to rate how well each of Hazan and Shaver's (1987) romantic attachment prototypes described them, it was revealed that the three ratings could be reduced to two dimensions, one corresponding to Avoidance (discomfort with closeness and dependency) and the other to Anxiety (about abandonment). In subsequent studies, Simpson (1990) and Collins and Read (1990) broke Hazan and Shaver's multi-sentence attachment-style prototypes into separate propositions with which subjects could agree or disagree to varying extents. When these Likert-type items were factor analyzed, a two-factor (Simpson) or three-factor (Collins & Read) solution was obtained. In the case of the three-factor solution, two of the factors (discomfort with closeness and discomfort with dependence on romantic partners) were significantly correlated (r = .38). Simpson and his colleagues (e.g., Simpson, Rholes, & Nelligan, 1992) called their two dimensions "security vs. avoidance" and "anxiety" (about abandonment). Collins and Read (1990) called their three dimensions "close," "depend," and "anxiety" (about abandonment). If we interpret the close and depend dimensions as facets of avoidance (the term facets being borrowed from Costa & McCrae, 1992), all of the early analyses of the structure of Hazan and Shaver's measure are compatible with the interpretation that adult attachment measures, like Ainsworth et al.'s coding scales for the Strange Situation, primarily assess avoidance and attachment-related anxiety.
The two-dimensional empirical and conceptual structure underlying attachment orientations was articulated more completely when researchers who study infant-caregiver attachment and those who study adolescent and adult romantic attachment realized that a two-dimensional space makes room for four, rather than three, quadrants or conceptual patterns. Crittenden (1988) and others who focused on infant-caregiver attachment in abusive and troubled families noted a mixed avoidant/anxious type. Main and Solomon (1990) identified a somewhat similar pattern, called "disorganized, disoriented" attachment. A diagram of the four infant types organized by the Avoidance and Anxiety dimensions is shown in Figure 2.
In the area of adult attachment, Bartholomew (1990), who had noticed that Hazan and Shaver's (1987, 1990) avoidant type and Main et al.'s (1985) dismissing (avoidant) type differed in the degree to which they exhibited anxious as well as avoidant qualities, proposed the now-familiar two-dimensional, four-category conceptual scheme shown in Figure 3. The parallels between Figures 2 and 3 are obvious. In both diagrams the upper left-hand quadrant represents securely attached individuals--infants and adults who are neither anxious about abandonment nor avoidant in their behavior. The upper right-hand quadrant of both diagrams represents anxious or preoccupied attachment, defined as a mixture of anxiety and interpersonal approach (nonavoidance). The lower left-hand quadrant represents dismissingly avoidant attachment, a combination of avoidant behavior and apparent lack of anxiety about abandonment. The lower right-hand quadrant represents fearfully avoidant attachment, which combines anxiety about abandonment with avoidant behavior.
----------------------------------------------
Insert Figures 2 and 3 about here,
preferably within sight of each other
----------------------------------------------
In subsequent work, Bartholomew has shown that it is possible to assess the four types and/or the two dimensions in adolescent and adult populations using either questionnaires or coded interviews (e.g., Bartholomew & Horowitz, 1991; Griffin & Bartholomew, 1994; Scharfe & Bartholomew, 1994). As shown in Figure 3, Bartholomew labels the two dimensions "model of self" and "model of other," but she and her coauthors also sometimes use the terms "anxiety" and "avoidance" (e.g., Scharfe, 1996), suggesting that a negative model of self is closely associated with anxiety about abandonment and that a negative model of others is closely associated with avoidant behavior. (Whether particular cognitive models, discussed in the present volume by Klohnen and John, actually account for the anxious and avoidant reactions or get built up around them during cognitive and personality development remains to determined.)
While the two-dimensional structure underlying adult attachment styles was being revealed, many researchers created their own measures, some in an attempt to tap the two dimensions (e.g., Wagner & Vaux, 1994) or the four styles defined by them (Griffin & Bartholomew, 1994). Others delineated additional styles (Hatfield, 1993; Latty-Mann & Davis, 1996) or included additional psychological content (e.g., anger; Sperling, Berman, & Fagan, 1994). Still others returned to Bowlby's (e.g., West & Sheldon-Keller, 1994) or Ainsworth's (e.g., Feeney, Noller, & Hanrahan, 1994) more specific constructs, such as compulsive self-reliance and separation protest, from which two or more dimensions might be constructed. Although each of these efforts made sense and yielded some interesting results, when first encountered en masse they constitute a bewildering obstacle to researchers who wish to study romantic attachment.
In the remainder of this chapter we will report some of the results of a large-sample study that incorporated most of the extant self-report attachment measures, including some that are rarely referenced by attachment researchers. We began with a thorough search of the literature, including available conference papers, from which we created a pool of 482 items designed to assess 60 named attachment-related constructs. The three of us then independently evaluated the degree of redundancy among similar items, reducing them to a single exemplary item if two or three of us agreed that they were completely or almost completely redundant. (As will be seen, this still left a substantial amount of inter-item similarity.) We thus reduced the 482 items to 323, from which all 60 subscale scores could be computed. We then factor-analyzed the 60 subscale scores, producing two essentially independent factors that correspond to the already-familiar Avoidance and Anxiety dimensions. When we clustered subjects into four groups based on their scores on the two factors, the groups corresponded conceptually to Bartholomew's four types (see our Figure 3). But the relations between the clusters and other theoretically appropriate target variables proved to be stronger than the corresponding relations between Bartholomew's self-report measure and those same target variables. We also computed two internally consistent but relatively brief scales to represent the Avoidance and Anxiety factors and used those scales to predict theoretically appropriate target variables. The results were promising and suggest that self-report attachment research might benefit from the use of the two scales. We turn now to a more detailed description of the study.
Combining All Self-Report Attachment Measures in a Single Questionnaire
Participants and Procedure
In order to perform reliable factor analyses with large numbers of items and constructs, we administered questionnaires to a sizable group of research subjects: 1,086 undergraduates, 682 women and 403 men, enrolled in psychology courses at the University of Texas at Austin. These students ranged in age from 16 to 50, with a median age of 18. Just under half the sample (487) described themselves as seriously involved in a relationship at the time of testing; the rest were dating casually (220) or not at all (376). Of those in a relationship, median relationship length was 15 months. Students received research credits in their classes for participating in the study, but their answers were completely anonymous. The questionnaire took approximately two hours to complete.
Materials
Attachment measures. The first set of five measures asked the students to classify themselves into one of three or four briefly described attachment-style categories. (The measures were those designed by Hazan and Shaver, 1987, 1990; Bartholomew and Horowitz, 1991; Sperling et al., 1994; and Latty-Mann and Davis, 1996.) For purposes of the present chapter, we will discuss data from only one categorical instrument, an adaptation of the Bartholomew and Horowitz (1991) measure that focused on experiences in romantic relationships.1
The second set of attachment measures included every multi-item scale of which we are aware, including some from never-published but useful conference papers. The items varied in content but all dealt with specific aspects of adolescent and adult attachment. Where necessary, we adapted item wording to emphasize romantic relationships (our own special interest) rather than all close relationships. After eliminating duplicate or very similar items (from different authors' scales), we were left with 323 statements that could be combined into a single questionnaire. The 323 items were printed in a randomly determined order. Space limitations here preclude a detailed discussion of each measure, but in general the following aspects of attachment were addressed: trust, separation protest, ambivalence, caregiving, careseeking, comfort with closeness, communication, commitment, avoidance, perceived partner availability, anxious attachment, alienation, angry withdrawal, loneliness, confidence in self and partner, defensiveness, disclosure, fear of rejection, jealousy/fear of abandonment, feared loss, proximity-seeking, self-reliance, viewing relationships as secondary, and romantic obsession. Subjects were asked to rate all 323 items on a 7-point scale ranging from "not at all like me" to "very much like me," a task requiring approximately 60 minutes. The following sources were drawn upon for items: Armsden and Greenberg (1987); Griffin and Bartholomew (1994); Brennan and Shaver (1995); Carnelley, Pietromonaco, and Jaffe (1994); Carver (1994); Collins and Read (1994); Feeney, Noller, and Hanrahan (1994); Hindy, Schwartz, and Brodsky (1989); Onishi and Gjerde (1993); Rothbard, Roberts, Leonard, and Eiden (1993); Shaver (1995); Simpson (1990); Wagner and Vaux (1994); West and Sheldon-Keller (1994).
Measures of personality and social behavior. Two additional kinds of measures were included in the study so that we could assess relations between self-report attachment constructs and two theoretically associated variables, intimate touch and romantic sexuality. Touch is an issue that has been addressed in studies of infant attachment (e.g., Main, 1990) but until recently not in studies of adult romantic attachment (see Brennan, Wu, & Loev, this volume). Sexuality was postulated by Shaver, Hazan, and Bradshaw (1988) to be one of three behavioral systems combined to form romantic love (along with attachment and caregiving). Except for a seminal study by Hazan, Zeifman, and Middleton (1994), however, sexuality has not been closely linked empirically with attachment patterns. The touch and sexuality measures used in the present study allowed us to determine how well various attachment measures fit within an interesting and under-studied nomological network of other variables. (Actually, we included several additional nodes in the nomological network but do not have space to discuss them here.) Each domain of questions will be described briefly.
Touch scales (Brennan et al., this volume). These 51 items assessed individual differences in touch within the context of romantic relationships. Of the seven constructs measured by the items, we will consider only four here: using touch to maintain affectionate proximity, desiring more physical contact, touch aversion, and using touch to assure a haven of safety.
Sex questions (Hazan et al., 1994; Janus & Janus, 1993; Laumann, Gagnon, Michael, & Michaels, 1994). Of the 47 items included in this domain, 21 assessed the degree to which subjects enjoy various kinds of sexual behavior on 7-point scales ranging ...
This is a preview of the whole essay
Touch scales (Brennan et al., this volume). These 51 items assessed individual differences in touch within the context of romantic relationships. Of the seven constructs measured by the items, we will consider only four here: using touch to maintain affectionate proximity, desiring more physical contact, touch aversion, and using touch to assure a haven of safety.
Sex questions (Hazan et al., 1994; Janus & Janus, 1993; Laumann, Gagnon, Michael, & Michaels, 1994). Of the 47 items included in this domain, 21 assessed the degree to which subjects enjoy various kinds of sexual behavior on 7-point scales ranging from "not at all" to "a great deal," and 12 assessed the frequency with which respondents experienced various emotions following sexual activities (e.g., feeling loved, sad, wanted/needed). For purposes of the present chapter, we will ignore the remaining items.
A principal components analysis of the 21 sexual-preference items produced five oblique factors, together accounting for 61.4% of the variance. The first of the resulting five unit-weighted scales assesses preference for "promiscuous" sexual behavior (e.g., one-night stands, alpha = .85). The second scale assesses what might be called "normative" sexual behavior in our sample (e.g., oral or manual stimulation of partners' genitals [two items], partners' oral/manual stimulation of subjects' genitals [two items], and vaginal intercourse; alpha = .87). The third scale assesses preference for sadistic or masochistic sexual behavior (alpha = .89). The fourth scale assesses preference for miscellaneous "non-normative" sexual behaviors (e.g., voyeurism, group sex, exhibitionism, attempting sexual contact with a nonconsenting person, anal intercourse; alpha = .71). The fifth scale assesses a preference for romantic/affectionate sexual behavior (cuddling, kissing, gazing; alpha = .66). For purposes of the present chapter we will consider only the three scales most relevant to attachment: promiscuous, normative, and affectionate sexuality.
Based on the same analytic procedures, the items assessing emotions typically experienced after sex, reasons for engaging in sexual behavior, and sexual problems all produced two-factor solutions. The 12 emotion items, which are of special interest to us in the present chapter, reduced to a "positive" factor (satisfied, loved, wanted/needed, taken care of, thrilled/excited) and a "negative" factor (scared/afraid, guilty, sad, anxious/worried, disgusted, frustrated/angry, alienated/lonely), which together accounted for 60.9% of the variance. Alphas was .87 for both scales.
Results of the Survey
Underlying Structure of the Attachment Measures
Correlations among 60 attachment constructs. We are unable to include the entire 60 X 60 matrix here, but we can summarize it briefly. Correlations among the 60 subscales were highly patterned: 62% were either > .50 (i.e., quite high) or < .20 (i.e., quite low). This pattern of high and low correlations suggested that there were in fact a few underlying factors to be identified, some of which were likely to be independent. Interestingly, the high correlations included ones spanning domains of measurement that have remained segregated in the literature. Table 2, for example, shows strong correlations between the subscales of Collins and Read's (1990), Armsden and Greenberg's (1987), and West and Sheldon-Keller's (1994) measures. These correlations are especially impressive given that each measure was developed under a different conceptual rubric and for a different research purpose. Collins and Read's measure was based on Hazan and Shaver's (1987) speculative extension of Ainsworth's infant typology to the realm of romantic love; Armsden and Greenberg's (1987) measure was designed to tap affective and cognitive dimensions of adolescents' attachment to peers; the subscales from West and Sheldon-Keller's measure were rooted more in Bowlby's clinical constructs than in Ainsworth's observational studies of infants. Notice that West and Sheldon-Keller's scales, which are not much discussed in the literature on romantic relationships, appear highly redundant with scales designed to measure romantic attachment--at least when they are completed with romantic attachment figures in mind. Despite the somewhat different aims and ideas behind these three sets of subscales, some of the correlations across the sets are quite high, indicating the existence of common underlying dimensions.
----------------------------------------------
Insert Table 2 about here
----------------------------------------------
Principal components analysis of the 60 attachment scales. A principal components analysis of the 60 subscales produced two major factors, which were rotated using an oblique procedure.2 Although this procedure allowed the factors to correlate with each other to any degree, the correlation between the two major factors was only .12, suggesting that the dimensions underlying attachment styles are essentially orthogonal. Together, the two factors accounted for 62.8% of the variance in the 60 subscales and were easily identifiable as 45-degree rotations of the dimensions obtained in previous work by Brennan, Shaver, and Tobey (1991) and Brennan and Shaver (1995). The two factors are conceptually equivalent to the horizontal and vertical axes of Bartholomew's four-category typology of attachment styles (e.g., Bartholomew & Horowitz, 1991; Griffin & Bartholomew, 1994). (Refer back to Figure 3.)
Table 3 presents the loadings of each of the 60 attachment subscales on the two higher-order factors. Avoidance of Intimacy (Rothbard et al., 1993), Discomfort with Closeness (Feeney et al., 1994), and Self-Reliance (West & Sheldon-Keller, 1994) emerged as the top three scales representative of the first factor, which we call Avoidance. Preoccupation (Feeney et al., 1994), Jealousy/Fear of Abandonment (Brennan & Shaver, 1995), and Fear of Rejection (Rothbard et al., 1993) emerged as the three scales most representative of the second factor, Anxiety.
----------------------------------------------
Insert Table 3 about here
----------------------------------------------
Creation of two higher-order scales. Two 18-item scales were constructed from the 36 items (out of the total pool of 323) with the highest absolute-value correlations with one of the two higher-order factors. Correlations of the two new (unit-weighted) scales, Avoidance and Anxiety, with the 60 major subscales and the four Bartholomew romantic style self-ratings are listed in Table 4. (The two scales, like their parent factors, are almost uncorrelated, r = .11; and each correlates very highly with its parent factor: r = .95 in both cases.) The correlations in Table 4 will allow experienced attachment researchers to see how their favorite scales fit into our two-dimensional scheme, and hopefully will help novices decide whether or not to use our scales for their particular purposes. Our Avoidance scale correlates highly with several other scales measuring avoidance and discomfort with closeness. The Anxiety scale correlates highly with scales measuring anxiety and preoccupation with attachment, jealousy, and fear of rejection.
----------------------------------------------
Insert Table 4 about here
----------------------------------------------
Creation of four clusters from the two higher-order scales. The two higher-order factors were used to cluster subjects according to guidelines suggested by Hair, Anderson, Tatham, and Black (1995, pp. 437-442). We first used a hierarchical clustering procedure (Ward's method, with squared Euclidean distance) to obtain initial cluster centers. Those centers were then used as a starting point for a second, nonhierarchical cluster analysis (K-Means, with an optimization method of assigning cases to clusters). We chose to conduct two cluster analyses, first hierarchical and then nonhierarchical, because nonhierarchical analyses typically provide more robust solutions. Nonhierarchical analyses allow cases to be switched from their initial cluster to a better-fitting cluster, a process known as optimizing, or "updating," the cluster centers. In hierarchical analyses, once a case has been assigned to a cluster center, it cannot be reassigned in a later iteration when alternative, better-fitting clusters may emerge. Here, the hierarchical cluster analysis was conducted first to provide initial cluster centers without which the nonhierarchical method would have had to use random starting points.
The initial pattern of clusters derived from Ward's hierarchical cluster analysis (in SPSS) revealed four distinct groups whose patterns of scores on the Avoidance and Anxiety factors clearly resembled Bartholomew's descriptions of the Secure, Fearful, Preoccupied, and Dismissing categories. Subjects in the 'Secure' cluster scored low on both Avoidance and Anxiety. Those in the 'Fearful' cluster scored high on both Avoidance and Anxiety. Those in the 'Preoccupied' cluster appeared low on Avoidance and high on Anxiety, while those in the 'Dismissing' cluster scored high on Avoidance and low on Anxiety. (The single quotation marks are intended to distinguish the four cluster-based styles from the similar styles derived from Bartholomew's measure.) Interestingly, when we instructed the program to find three clusters, the two avoidant clusters were collapsed into one, similar to the way Hazan and Shaver (1987) originally measured romantic attachment. When five clusters were selected, the preoccupied group split into those very low on Avoidance but only moderately high on Anxiety and those very high on Anxiety but only moderately low on Avoidance, a distinction not previously made in the romantic attachment literature. (This split may have been encouraged by the larger-than-usual number of items on our scales. As explained later, we do not wish to place much emphasis on any particular number of types because other analyses, reported in the present volume by Fraley and Waller, indicate that the types are not "real" in any case. Ultimately, only the dimensional scores matter.) For purposes of comparing scale-based clusters with the types generated by Bartholomew's self-report measure, we chose the four-cluster solution, which is justified by the pattern of agglomeration coefficients, the dendrogram, and previous research and theory. (Cf. Feeney et al., 1994, for similar analyses and a similar conclusion.)
The pattern of clusters from the second, nonhierarchical cluster analysis was quite similar to that from the hierarchical analysis. A look at the final cluster centers revealed that the 'Secure' cluster center was low on both Avoidance (-1.02) and Anxiety (-.67) factors, whereas the 'Fearful' cluster was high on both Avoidance (.94) and Anxiety (.52). The 'Preoccupied' cluster was low on Avoidance (-.34) and high on Anxiety (1.10). The 'Dismissing' cluster was high on Avoidance (.78) and low on Anxiety (-.91). This cluster-based attachment-style category system will be used in subsequent analyses in this chapter, beginning with a detailed comparison between it and Bartholomew's self-report measure (Bartholomew & Horowitz, 1991, modified to refer to romantic relationships). Readers who wish to use our scales will find the questionnaire in Appendix A and scoring instructions in Appendix B. It is not necessary to ask us for permission to use the scales.
Comparison of Categorical Attachment Measures
Association with each other. Table 5 lists the cross-tabulation of attachment style categories assessed with the Bartholomew four-category self-classification measure and our new, cluster-based method. A chi-square test comparing the two assessment schemes was highly significant, indicating substantial similarity. Collapsing across secure and insecure categories allows meaningful comparison of the two measures' rates of secure vs. insecure classifications. Just over half (52.8%) of the people classified as Secure on Bartholomew's measure were placed in the 'Secure' category of the new, cluster-based classification. Only a very small minority (11.2%) of those classified as insecure on Bartholomew's measure were classified as 'Secure' on the new measure. A very large percentage (88.8%) of the sample classified as insecure with Bartholomew's measure were also classified as insecure by the new measure. But nearly half (47.2%) classified as Secure on Bartholomew's measure were classified as insecure on the new measure. In other words, subjects were more likely to be categorized as insecure and less likely to be categorized as Secure using our new measure compared with Bartholomew's measure, so the new procedure is more conservative than Bartholomew's in classifying a person as secure.3 This is probably because our scales discriminate more precisely than Bartholomew's measure among people with different degrees of insecurity. At least in relation to the nomological network we have been exploring in our research, this conservatism about calling a person secure generally leads to statistically stronger results, as illustrated in the next two sections.
----------------------------------------------
Insert Table 5 about here
----------------------------------------------
Associations between Bartholomew's measure, Anxiety and Avoidance, and touch and sex subscales. Three multivariate analyses of variance (MANOVAs) were computed on (1) the Anxiety and Avoidance scales, (2) the touch subscales, and (3) the sex subscales as a function of the Bartholomew self-classification measure. The first MANOVA on the two higher-order factor scales was highly significant, as were the MANOVAs on the touch and sex subscales. The means for the two higher-order scales and the touch and sex subscales are listed in Table 6, along with the univariate Fs. Follow-up Tukey pairwise comparisons were computed to determine which groups differed significantly.
----------------------------------------------
Insert Table 6 about here
----------------------------------------------
Both the Fearful and Dismissing groups, which did not differ from each other, scored higher on the Avoidance scale than either the Secure or the Preoccupied group; Preoccupied individuals scored higher than Secures. Both Secure and Dismissing individuals scored lower on Anxiety than either Fearful or Preoccupied individuals; Preoccupied individuals scored even higher than Fearfuls on the Anxiety scale. These findings suggest, as expected, that Anxiety is similar to Bartholomew's self-model dimension, and Avoidance is similar to her other-model dimension. (See also Griffin & Bartholomew, 1994, and Simpson, Rholes, & Phillips, 1996.)
Bartholomew's Secure and Preoccupied groups scored similarly on the touch subscales. Both groups scored high on using touch to maintain affectionate proximity. Both scored low on the touch aversion scale. The two kinds of avoidants were similarly touch averse. Both avoidant groups reported a moderate amount of touch deficit in their relationships (suggesting that they are not oblivious to such deficits). Preoccupied individuals were most different from Dismissing individuals in using touch to seek care (establish a safe haven) in their relationships. (Dismissing individuals are what Bowlby and many subsequent attachment researchers have called "compulsively self-reliant.") Safe-haven touch was the only scale on which Fearful and Secure subjects scored similarly--in between Preoccupied and Dismissing subjects. Finally, Preoccupied individuals distinguished themselves by desiring more touch than they were receiving. All of these findings are compatible with previous research and theoretical writings on adult attachment.
In terms of preferences for various sexual behaviors, interesting differences emerged among the four Bartholomew attachment categories. In general, Secures, along with Preoccupieds, were most likely to endorse romantic/affectionate sexual behaviors. Secures also differed from Fearful and Dismissing individuals in their preference for "normative" sexual behaviors. Of the four groups, Dismissing individuals were the most likely to endorse promiscuous sexual behavior.
Secures were also more likely than insecures to experience positive emotions, and less likely to experience negative emotions, following sex. Fearful individuals scored highest on the negative emotions scale, followed by Preoccupied and then Dismissing individuals.
Associations between the new attachment-categories, Anxiety and Avoidance, and the touch and sex subscales. Three MANOVAs, paralleling the ones just described, were computed on (1) the Anxiety and Avoidance scales, (2) the touch subscales, and (3) the sex subscales as a function of the new cluster-based attachment categories. The first MANOVA on the two higher-order scales was highly significant, as were the MANOVAs on the touch and sex subscales. All of the F values were much higher than the corresponding values based on Bartholomew's categorical self-report measure. The attachment-group means for the two higher-order scales and all of the touch and sex subscales are listed in Table 7, along with the univariate Fs. Follow-up Tukey pairwise comparisons were computed to determine which groups differed significantly.
----------------------------------------------
Insert Table 7 about here
----------------------------------------------
The findings for the univariate Avoidance, Anxiety, touch, and sex scales were similar to those obtained with Bartholomew's measure, but the F values were all substantially higher with the new scale-based clusters. (Compare corresponding sections of Tables 6 and 7--particularly the column showing variance accounted for [?2].) Also, the pairwise group differences were sharper with the new measure. For example, on the safe-haven touch subscale, all four attachment groups differed using the new measure, whereas Secure and Fearful groups were indistinguishable when Bartholomew's measure was used.
We also conducted a series of simultaneous regression analyses to examine the ability of the two dimensions, Avoidance and Anxiety, to account for variance in the touch and sex scores. Table 8 displays the parameter estimates (?2s) for predicting the sex and touch scores from the Bartholomew and the new cluster-based measures as well as from the continuous Avoidance and Anxiety scales (R2s). This time we included the interaction effects produced when each categorical measure was treated as two separate variables--model of self and model of other--as well as the interaction term created by multiplying the Avoidance and Anxiety scores.4 (The interaction terms for the categorical measures did not generally produce large effects, which is why the overall parameter estimates for the ANOVAs in Table 8 are not very different from those reported in Tables 6 and 7.) Notice in Table 8 that the variance accounted for, on average, by the new categorical measure is greater than two and a half times the variance accounted for by Bartholomew's measure. Furthermore, the variance accounted for by the Avoidance and Anxiety scales is over three and a half times as high as that accounted for by Bartholomew's measure, and about a third higher than that accounted for by the new cluster-based categories. This pattern indicates that the dimensions are more powerful than the categories, a conclusion also reached by Fraley and Waller (this volume).
----------------------------------------------
Insert Table 8 about here
----------------------------------------------
Finally, we should add that, when factored at the item level, our pool of 323 items produced 12 specific attachment-related dimensions, each with enough high-loading items (viz. 10) to produce reliable unit-weighted scales. Not surprisingly, a higher-order factor analysis of the 12 scales again revealed two underlying dimensions which, when rotated, corresponded to the familiar Avoidance and Anxiety constructs. The 12 scales, which can be viewed as "facets" (Costa & McCrae, 1992) of Avoidance and Anxiety, are as follows: (1) Partner is a Good Attachment Figure; (2) Separation Anxiety; (3) Self-Reliance; (4) Discomfort with Closeness; (5) Attachment-Related Anger at Partners; (6) Uncertainty About Feelings for Partners; (7) Discomfort with Dependence; (8) Trust; (9) Lovability/Relational Self-Esteem; (10) Desire to Merge with Partners; (11) Tough-Minded Independence; and (12) Fear of Abandonment. These scales are treated in more detail elsewhere (Brennan, Clark, & Shaver, 1996). Here, we mainly want to remind readers that romantic attachment researchers have been measuring a fairly broad array of attachment-related constructs, some of which may be worth highlighting in particular future studies. At a more abstract level, however, these lower-order constructs reduce to the two dimensions of Avoidance and Anxiety.
Implications for the Measurement of Adult Romantic Attachment
In writing this chapter, we had three purposes in mind. First, we sought to provide a concise overview of issues related to the self-report measurement of adult romantic attachment. Second, we sought to encourage researchers to use a common metric for assessing adult romantic attachment styles. Toward this goal, we offer our two attachment-dimension scales as useful tools for researchers hoping to circumvent the unreliability inherent in single-item response formats. Our scales, Avoidance and Anxiety, have the advantage of being derived from virtually every other extant self-report adult romantic attachment measure. That is, these two dimensions underlie virtually all self-report adult romantic attachment measures and appear crucial for capturing important individual differences in adult romantic attachment. Of course, we are claiming no originality. Our scales are based on other researchers' items (many of them adapted from Hazan & Shaver, 1987), and our conclusions are compatible with those of Simpson (1990), Bartholomew and Horowitz (1991), and others. But our two 18-item scales, having high internal consistency, and being based on a large, comprehensive item pool, may be more precise than previous scales. The two dimensions have the advantage of being analogous to the ones first discovered by Ainsworth and her colleagues (Ainsworth et al., 1978). (As explained earlier in this chapter, Ainsworth et al. were able to distinguish among secure, avoidant, and anxious-ambivalent babies using two discriminant functions, similar to our Avoidance and Anxiety dimensions, formed by the continuous scales used to code infants' behavior in the Strange Situation.)
Third, we sought to define attachment patterns, or styles, in terms of regions in a two-dimensional space. Our two 18-item attachment scales can be used to classify individuals into one of four adult romantic attachment categories. This categorization procedure produces stronger results than Bartholomew's self-classification measure, at least in relation to measures of attachment-related emotions, thoughts, and behaviors regarding touch and sexuality in romantic relationships. We are conducting additional comparisons of the two classification procedures in studies of other topics of interest to attachment researchers--for example, partner abuse, violence in one's family of origin, and relationship-initiation strategies. We want to reiterate, however, that categorization of research participants is unnecessary when dimensional measures are available; and some power and precision are lost when categories rather than continuous scales are used. Given that Fraley and Waller (this volume) found no evidence for the categorical nature of attachment styles, it is difficult to justify categorical measures except on grounds of convenience.
In this chapter we deliberately side-stepped the issue of self-report versus interview measures of attachment style. Researchers who use interview measures (e.g., Bartholomew & Horowitz, 1991; Main et al., 1985) generally believe that interviews are more powerful and revealing than self-report measures. There are good reasons why they may be correct. Self-report measures are subject to response biases, and they rely on research participants' honesty and self-insight, which are probably limited in any case but especially so when fears and defenses are at issue. It remains to be seen how well the dimensional measures described here stand up in comparisons with interview measures. We think it is possible that the increased precision we have demonstrated in relation to Bartholomew's self-classification measure will make our dimensional measures more similar to interviews. (See Bartholomew & Shaver, this volume, for a discussion of the importance of precision and statistical power in revealing weak but valid associations.) And our longer-than-usual scales may circumvent some of the temptation toward biased responding aroused by simple measures that require people to say fairly directly whether they are or are not secure. It is worth noting that self-report measures of attachment, like interview measures, do not require that people understand or probe into their own dynamics and defenses. Self-report measures require only a modicum of familiarity with one's own feelings, social behavior, and beliefs about relationships and the feedback one has received from relationship partners. It is possible to classify people on these grounds without them understanding their own histories or dynamics.
Looking ahead to the next decade of research on romantic attachment, we are cautiously optimistic. We are cautious because of the field's continuing lack of convergence on a common, reliable method for assessing adult attachment orientations. A common method is necessary if researchers are to communicate clearly with each other about the same constructs, if neophyte researchers are to enter the field relatively easily and quickly, and if all researchers are to move on to substantive issues rather than remaining hung up on psychometric ones. We are optimistic because of the remarkable strides made in the previous decade and because there is more commonality underlying different research procedures than might have been expected. We have shown here, in line with previous work by Simpson, Bartholomew, and their coauthors, that everyone is working with the same two dimensions that Ainsworth and her colleagues identified in 1978: Avoidance and Anxiety. The origins and implications of people's scores on those dimensions are what all attachment researchers deal with, whether knowingly or not.
References
Ainsworth, M. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Patterns of attachment: A psychological study of the Strange Situation. Hillsdale, NJ: Erlbaum.
Armsden, G. C., & Greenberg, M. T. (1987). The inventory of parent and peer attachment: Individual differences and their relationship to psychological well-being in adolescence. Journal of Youth and Adolescence, 16, 427-454.
Bartholomew, K. (1990). Avoidance of intimacy: An attachment perspective. Journal of Social and Personal Relationships, 7, 147-178.
Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles among young adults: A test for a four-category model. Journal of Personality and Social Psychology, 61, 226-244.
Bowlby, J. (1969/1982). Attachment, Vol. 1 of Attachment and loss (2nd ed.). New York: Basic Books.
Brennan, K. A., & Shaver, P. R. (1995). Dimensions of adult attachment, affect regulation, and romantic relationship functioning. Personality and Social Psychology Bulletin, 21, 267-283.
Brennan, K. A., Shaver, P. R., & Tobey, A. E. (1991). Attachment styles, gender, and parental problem drinking. Journal of Social and Personal Relationships, 8, 451-466.
Carnelley, K., Pietromonaco, P., & Jaffe, K. (1994). Depression, working models of others, and relationship functioning. Journal of Personality and Social Psychology, 66, 127-140.
Carver, C. S. (1994). Personality and adult attachment. Unpublished manuscript, University of Miami, Coral Gables, FL.
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.
Costa, P. T., Jr., & McCrae, R. (1992). Revised NEO Personality Inventory: NEO-PI and NEO Five-Factor Inventory (Professional Manual). Odessa, FL: Psychological Assessment Resources.
Crittenden, P. M. (1988). Relationships at risk. In J. Belsky & T. Nezworski (Eds.), Clinical implications of attachment (pp. 136-174). Hillsdale, NJ: Erlbaum.
Feeney, J. A., Noller, P., & Hanrahan, M. (1994). Assessing adult attachment. In M. B. Sperling and W. H. Berman (Eds.), Attachment in adults: Clinical and developmental perspectives (pp. 128-152). New York: Guilford.
George, C., Kaplan, N., & Main, M. (1985). Adult Attachment Interview. Unpublished manuscript, Department of Psychology, University of California, Berkeley, CA.
Griffin, D., & Bartholomew, K. (1994). Models of the self and other: Fundamental dimensions underlying measures of adult attachment. Journal of Personality and Social Psychology, 67, 430-445.
Hair, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (1995). Multivariate data analysis, with readings (4th ed.). Englewood Cliffs, NJ: Prentice-Hall.
Hatfield, E. (1993, October). Varieties of romantic attachment. Paper presented at the annual meeting of the Society of Experimental Social Psychology, Santa Barbara, CA.
Hazan, C., & Shaver, P. (1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 52, 511-524.
Hazan, C., & Shaver, P. R. (1990). Love and work: An attachment-theoretical perspective. Journal of Personality and Social Psychology, 59, 270-280.
Hazan, C., Zeifman, D., & Middleton, K. (1994, July). Adult romantic attachment, affection, and sex. Paper presented at the Seventh International Conference on Personal Relationships, Groningen, The Netherlands.
Hindy, C. G., Schwartz, J. C., & Brodsky, A. (1989). If this is love, why do I feel so insecure? New York: Atlantic Monthly Press.
Janus, S. S., & Janus, C. L. (1993). The Janus report on sexual behavior. New York: Wiley and Sons.
Latty-Mann, H., & Davis, K. E. (1996). Attachment theory and partner choice: Preference and actuality. Journal of Social and Personal Relationships, 13, 5-23.
Laumann, E. O., Gagnon, J. H., Michael, R. T., & Michaels, S. (1994). The social organization of sexuality: Sexual practices in the United States. Chicago, IL: The University of Chicago Press.
Levy, M. B., & Davis, K. E. (1988). Lovestyles and attachment styles compared: Their relations to each other and to various relationship characteristics. Journal of Social and Personal Relationships, 5, 439-471.
Main, M. (1990). Parental aversion to infant-initiated contact is correlated with the parent's own rejection during childhood: The effects of experience on signals of security with respect to attachment. In K. E. Barnard & T. B. Brazelton (Eds.), Touch: The foundation of experience (pp. 461-495). Madison, CT: International Universities Press.
Main, M., Kaplan, N., & Cassidy, J. (1985). Security in infancy, childhood, and adulthood: a move to the level of representation. In I. Bretherton & E. Waters (Eds.), Growing points in attachment theory and research. Monographs of the Society for Research in Child Development, 50 (1-2, Serial No. 209), 66-106.
Main, M., & Solomon, J. (1990). Procedures for identifying insecure-disorganized/disoriented infants: Procedures, findings, and implications for the classification of behavior. In M. Greenberg, D. Cicchetti, & M. Cummings (Eds.), Attachment in the preschool years: Theory, research, and intervention (pp. 121-160). Chicago, IL: University of Chicago Press.
Onishi, M., & Gjerde, P. F. (1994). Adult attachment styles: A multi-method examination of personality characteristics. Presented at the biennial meeting of the International Society for the Study of Behavioral Development, Amsterdam, The Netherlands.
Rothbard, J. C., & Roberts, L. J., Leonard, K. E., & Eiden, R. D. (1993, August). Attachment styles and marital interaction behavior. Presented at the annual meeting of the American Psychological Association, Toronto, Ontario.
Scharfe, E. (1996). A test of Bartholomew's four-category model of attachment in a clinical sample of adolescents. Unpublished doctoral dissertation, Department of Psychology, Simon Fraser University, Burnaby, British Columbia.
Scharfe, E., & Bartholomew, K. (1994). Reliability and stability of adult attachment patterns. Personal Relationships, 1, 23-43.
Shaver, P. R. (1995). Additional items to measure dismissing avoidance. Unpublished technical memorandum, Department of Psychology, University of California, Davis.
Simpson, J. A. (1990). Influence of attachment styles on romantic relationships. Journal of Personality and Social Psychology, 59, 971-980.
Simpson, J. A., Rholes, W. S. , & Nelligan, J. S. (1992). Support-seeking and support-giving within couple members in an anxiety-provoking situation: The role of attachment styles. Journal of Personality and Social Psychology, 62, 434-446.
Simpson, J. A., Rholes, W. S. , & Phillips, D. (1996). Conflict in close relationships: An attachment perspective. Journal of Personality and Social Psychology, 71, 799-914.
Sperling, M. B., Berman, W. H., & Fagen, G. (1992). Classification of adult attachment: An integrative taxonomy from attachment and psychoanalytic theories. Journal of Personality Assessment, 59, 239-247.
van IJzendoorn, M. H. (1995). Adult attachment representations, parental responsiveness, and infant attachment: A meta-analysis on the predictive validity of the Adult Attachment Interview. Psychological Bulletin, 117, 387-403.
Wagner, L., & Vaux, A. (1994, May). The Wagner Attachment Scale (WAS): Reliability and validity of a self-report measure of adult attachment. Presented at the annual meeting of the Midwestern Psychological Association. Chicago, IL.
West, M., & Sheldon-Keller, A. E. (1994). Patterns of relating: An adult attachment perspective. NY: Guilford.
Footnotes
1We modified the instructions for the Bartholomew and Horowitz (1991) attachment measure to make it refer specifically to romantic relationships: "Read each of the four self-descriptions below and place a check mark next to the single alternative that best describes how you usually act and feel in romantic relationships or that comes nearest to describing you." In our close relationship version, the word "romantic" was replaced with "close." Both measures included the following four options: (Secure) It is easy for me to become emotionally close to others. I am comfortable depending on others and having others depend on me. I don't worry about being alone or having others not accept me. (Fearful) I am uncomfortable getting close to others. I want emotionally close relationships, but I find it difficult to trust others completely, or to depend on them. I worry that I will be hurt if I allow myself to become too close to others. (Preoccupied) I want to be completely emotionally intimate with others, but I often find that others are reluctant to get as close as I would like. I am uncomfortable being without close relationships, but I sometimes worry that others don't value me as much as I value them. (Dismissing) I am comfortable without close emotional relationships. It is very important to me to feel independent and self-sufficient, and I prefer not to depend on others or have others depend on me. (The romantic and peer measures were highly associated, ?2 [9] = 1049.07, p < .0001, as Bartholomew expected when she decided to use one wording to apply to all close relationships with peers, whether romantic or not.)
2We actually conducted the factor analyses twice, once in each of two halves of the sample. The results were virtually identical (the same two factors, with the same items loading on each factor in almost exactly the same order), so here we present only the results for the sample as a whole.
3The number of people who are "truly" secure or insecure is something that cannot be stated firmly. The Adult Attachment Interview (Main et al., 1985) tends to place around 65% of adults into the secure category because it was designed to predict the Strange Situation classification of subjects' 12-month-old children. And Ainsworth originally said that around 65% of middle-class American infants were secure (based on the scoring criteria she invented). Bartholomew's adaptation of the AAI procedure for assessing adult attachment to peers and romantic partners (Bartholomew & Horowitz, 1991) tends to label around 45% of college students as secure. This is simply a function of how she instructs coders to identify security. Hazan and Shaver's (1987, 1990) simple three-category measure results in around 55% of college students and non-student adults calling themselves secure. The cluster analysis reported here, based on scales whose items refer mainly to nuances of insecurity (Anxiety and Avoidance), places relatively few people in the secure category, with good effects on analyses to be reported later. But Fraley and Waller's chapter in the present volume suggests that all such category breaks are somewhat arbitrary, given that there are no clear or natural category boundaries between attachment styles.
4For each categorical measure, models of self and other were computed as follows. Model of self was coded as one for individuals with secure or dismissing attachment styles, and coded as zero for those with fearful or preoccupied attachment styles. Model of other was coded as one for those with secure or preoccupied attachment styles, and as zero for those with dismissing or fearful attachment styles. The two resulting variables--self and other--were then treated as grouping variables (with two levels each) in a two-way analysis of variance.
Table 1
Hazan & Shaver's (1987, 1990) Three Prototypes
Avoidant. I am somewhat uncomfortable being close to others; I find it difficult to trust them completely, difficult to allow myself to depend on them. I am nervous when anyone gets too close, and often, love partners want me to be more intimate than I feel comfortable being. Anxious-ambivalent. I find that others are reluctant to get as close as I would like. I often worry that my partner doesn't really love me or won't want to stay with me. I want to get very close to my partner, and this sometimes scares people away.
Secure. I find it relatively easy to get close to others and am comfortable depending on them. I don't often worry about being abandoned or about someone getting too close to me.
Table 2
Selected Examples of the Intercorrelations of the 60 Subscales
___________________________________________________________________________
Collins & Read's (1990) scales
______________________________
Anxious Close Depend
___________________________________________________________________________
Armsden & Greenberg's (1987) scales
_______________________________
Alienation .62 -.39 -.62
Trust -.50 .49 .66
Communication -.41 .49 .58
West & Sheldon-Keller's (1994) scales
________________________________
Angry withdrawal .57 -.18 -.51
Availability of partner -.61 .46 .73
Compulsive caregiving .08 .32 .18
Compulsive careseeking .19 .12 .04
Feared loss of partner .78 -.30 -.54
Proximity-seeking .18 .21 .06
Compulsive self-reliance .40 -.77 -.65
Separation protest .30 .05 -.12
Use partner as secure base -.30 .57 .58
___________________________________________________________________________
Note. All coefficients above .11 are statistically significant at p < .001.
Table 3
Factor Loadings of the 60 Attachment Subscales with Two Higher-Order Factors
___________________________________________________________________________
Attachment subscale Factor 1 Factor 2
___________________________________________________________________________
Avoidance of intimacy .91
(Rothbard et al., 1993)
Discomfort with closeness .90
(Feeney et al., 1994)
Self-reliance .88
(West & Sheldon-Keller, 1994)
Avoidance .86
(Carver, 1994)
Discomfort with closeness .86
(Carnelley et al., 1994)
Self-reliance .86
(Carnelley et al., 1994)
Fearful prototype .85
(Onishi & Gjerde, 1993)
Discomfort with disclosure .85
(Carnelley et al., 1994)
Comfort with closeness -.84
(Collins & Read, 1990)
Security -.83
(Simpson, 1990)
Proximity-seeking -.82
(Brennan & Shaver, 1995)
Dismissing prototype .82 -.40
(Onishi & Gjerde, 1993)
Self-reliance .82
(Brennan & Shaver, 1995)
Secure prototype -.82
(Onishi & Gjerde, 1993)
Fearfulness .82
(Bartholomew, 1994)
Table 3 (cont.)
Factor Loadings of the 60 Attachment Subscales with Two Higher-Order Factors
___________________________________________________________________________
Attachment subscale Factor 1 Factor 2
___________________________________________________________________________
Avoidance .81
(Simpson, 1990)
Use partner as secure base -.79
(West & Sheldon-Keller, 1994)
Comfort depending on others -.79
(Collins & Read, 1990)
Ambivalence .76
(Brennan & Shaver, 1995)
Ambivalence .75
(Carnelley et al., 1994)
Dismissiveness .73
(Shaver, 1995)
Defensiveness .72
(Carnelley et al., 1994)
Communication -.72
(Armsden & Greenburg, 1987)
Trust -.72
(Brennan & Shaver, 1995)
Trust -.72
(Armsden & Greenburg, 1987)
Model of others -.70 -.31
(Wagner & Vaux, 1994)
Security -.70
(Carver, 1994)
Self-confidence -.70
(Feeney et al., 1994)
Security -.69 -.30
(Bartholomew, 1994)
Distrust .68 .39
(Carnelley et al., 1994)
Table 3 (cont.)
Factor Loadings of the 60 Attachment Subscales with Two Higher-Order Factors
___________________________________________________________________________
Attachment subscale Factor 1 Factor 2
___________________________________________________________________________
Availability of partners -.66 -.39
(West & Sheldon-Keller, 1994)
Relationships as secondary .61
(Feeney et al., 1994)
Frustration with partners .60 .50
(Brennan & Shaver, 1995)
Model of self -.52 -.35
(Wagner & Vaux, 1994)
Dismissiveness .50 -.42
(Bartholomew, 1994)
Compulsive caregiving -.46 .45
(West & Sheldon-Keller, 1994)
Table 3 (cont.)
Factor Loadings of the 60 Attachment Subscales with Two Higher-Order Factors
___________________________________________________________________________
Attachment subscale Factor 1 Factor 2
___________________________________________________________________________
Preoccupation .86
(Feeney et al., 1994)
Jealousy/Fear of abandonment .85
(Brennan & Shaver, 1995)
Fear of rejection .83
(Rothbard et al., 1993)
Preoccupied prototype -.39 .82
(Onishi & Gjerde, 1993)
Jealousy .80
(Carnelley et al., 1994)
Preoccupation .79
(Bartholomew, 1994)
Anxiety .77
(Simpson, 1990)
Worry .76
(Carver, 1994)
Anxious-clinging to partners .33 .76
(Brennan & Shaver, 1995)
Anxious-clinging to partners .75
(Carnelley et al., 1994)
Anxiety .41 .74
(Collins & Read, 1990)
Anxious attachment .72
(Hindy et al., 1991)
Romantic anxiety .38 .71
(Hindy et al., 1991)
Proximity-seeking .68
(Carnelley et al., 1994)
Romantic obsession -.40 .67
(Hindy et al., 1991)
Table 3 (cont.)
Factor Loadings of the 60 Attachment Subscales with Two Higher-Order Factors
___________________________________________________________________________
Attachment subscale Factor 1 Factor 2
___________________________________________________________________________
Desire to merge with partner .66
(Carver, 1994)
Dependent self-esteem .66
(Rothbard et al., 1993)
Angry withdrawal .65
(West & Sheldon-Keller, 1994)
Feared loss of partner .41 .64
(West & Sheldon-Keller, 1994)
Need for approval .62
(Feeney et al., 1994)
Proximity-seeking -.36 .59
(West & Sheldon-Keller, 1994)
Compulsive careseeking .57
(West & Sheldon-Keller, 1994)
Alienation .53 .53
(Armsden & Greenburg, 1987)
Separation protest .48
(West & Sheldon-Keller, 1994)
___________________________________________________________________________
Note. Loadings lower than .30 were omitted.
Table 4
Correlations of 60 Attachment Subscales with Two Higher-Order Attachment Scales
___________________________________________________________________________
Avoidance Anxiety
___________________________________________________________________________
Armsden & Greenberg's (1987) scales
Alienation .51 .58
Communication -.68 -.24
Trust -.65 -.34
Bartholomew's (1994) scales
Dismissiveness .39 -.29
Fearfulness .81 .32
Preoccupation -.13 .73
Security -.70 -.46
Brennan & Shaver's (1995) scales
Ambivalence .73 .30
Anxious-clinging to partners .31 .78
Jealousy/Fear of abandonment .11 .82
Frustration with partners .54 .56
Proximity-seeking -.78 .12
Self-reliance .79 .00
Trust -.66 -.37
Carnelley et al.'s (1994) scales
Ambivalence .75 .26
Anxious-clinging to partners .02 .75
Defensiveness .68 .20
Discomfort with closeness .86 -.02
Discomfort with disclosure .86 .14
Distrust .62 .46
Jealousy .26 .80
Proximity-seeking -.05 .60
Table 4 (cont.)
Correlations of 60 Attachment Subscales with Two Higher-Order Attachment Scales
___________________________________________________________________________
Avoidance Anxiety
___________________________________________________________________________
Carver's (1994) scales
Avoidance .90 .09
Desire to merge with partners .16 .70
Security -.63 .14
Worry .27 .79
Collins & Read's (1990) scales
Anxiety .41 .79
Comfort with closeness -.87 -.05
Comfort depending on others -.73 -.39
Feeney et al.'s (1994) scales
Self-confidence -.69 -.25
Discomfort with closeness .88 .22
Need for approval .24 .63
Preoccupation .13 .88
Relationships as secondary .56 .16
Hindy et al.'s (1991) scales
Anxious attachment .22 .71
Romantic anxiety .35 .75
Romantic obsession -.34 .55
Onishi & Gjerde's (1993) scales
Dismissing prototype .73 -.22
Fearful prototype .80 .33
Preoccupied prototype -.34 .68
Secure prototype -.79 -.39
Rothbard et al.'s (1993) scales
Avoidance of intimacy .89 .07
Dependent self-esteem -.07 .60
Fear of rejection .29 .88
Shaver's (1995) scales
Dismissiveness .64 -.12
Table 4 (cont.)
Correlations of 60 Attachment Subscales with Two Higher-Order Attachment Scales
___________________________________________________________________________
Avoidance Anxiety
___________________________________________________________________________
Simpson's (1990) scales
Anxiety .06 .75
Avoidance .82 .20
Security -.84 -.28
Wagner & Vaux's (1994) scales
Model of self -.52 -.42
Model of others -.63 -.40
West & Sheldon-Keller's (1994) scales
Angry withdrawal .26 .67
Availability of partner -.61 -.47
Compulsive caregiving -.36 .30
Compulsive careseeking -.18 .49
Feared loss of partner .39 .69
Proximity-seeking -.28 .48
Compulsive self-reliance .88 .14
Separation protest -.02 .48
Use partner as secure base -.78 -.08
___________________________________________________________________________
Note. Scales are listed in alphabetical order by author and, within author, by scale name.
All coefficients above .11 are statistically significant at p < .001.
Table 5
Relationship Between Bartholomew's and Cluster-Based Attachment-Style Categories
_____________________________________________________________________________
Bartholomew's attachment-style category
Cluster-based Secure Fearful Preoccupied Dismissing Row Total
category
_____________________________________________________________________________
Secure 264 21 20 24 329 (30.4%)
Fearful 65 116 38 45 264 (24.4%)
Preoccupied 94 52 112 6 264 (24.4%)
Dismissing 77 60 4 84 225 (20.8%)
_____________________________________________________________________________
Column Total: 500 249 174 159 1082 (100%)
(46.2%) (23.0%) (16.1%) (14.7%)
_____________________________________________________________________________
Note. ?2 (df = 9) = 497.78, p < .0001; N = 1082.
Table 6
Two New Attachment-Scale Scores and Touch and Sex Subscale Scores as a Function of Bartholomew's
Attachment-Style Category
_____________________________________________________________________________________
Bartholomew's attachment-style category
Secure Fearful Preoccupied Dismissing Univariate F's ?2
New scales (df 3, 1073)
Avoidance 2.36a 3.71c 2.68b 3.74c 151.00*** .30
Anxiety 3.08a 3.83b 4.54c 2.99a 132.12*** .27
Touch (df 3, 1051)
Affectionate
touch 5.79c 5.47b 5.86c 4.98a 28.49*** .08
Desire for
more touch 2.45a 2.86b 3.30c 2.71b 27.21*** .07
Touch
aversion 2.15a 2.66b 2.24a 2.64b 17.61*** .05
Haven of
safety 4.48b 4.34b 4.96c 3.95a 23.57*** .06
Sexual preferences (df 3, 1055)
Promiscuous 1.91a 2.01a 2.06a 2.58b 9.70*** .03
Normative 5.49b 5.10a 5.40ab 5.13a 5.01** .01
Affectionate 6.11c 5.86b 6.08c 5.59a 13.87*** .04
Post-coital emotions (df 3, 698)
Positive 3.90a 3.35b 3.48b 3.20b 17.98*** .07
Negative .84a 1.38c 1.22bc 1.08b 15.71*** .06
_________________________________________________________________________________________
Note. Numbers in the first four columns are means. Means within each row whose superscripts differ are
different at p < .05. ** p < .01, *** p < .001 (two-tailed).
Table 7
Two New Attachment-Scale Scores and Touch and Sex Subscale Scores as a Function of the Cluster-Based Attachment-Style Category
_____________________________________________________________________________________
Cluster-based attachment-style category
Secure Fearful Preoccupied Dismissing Univariate F's ?2
New scales (df 3, 1076)
Avoidance 1.88a 3.96c 2.40b 3.87c 614.67*** .63
Anxiety 2.64a 4.06b 4.60c 2.60a 604.41*** .63
Touch (df 3, 1052)
Affectionate
touch 6.10c 5.14b 6.05c 4.93a 104.23*** .23
Desire for
more touch 2.07a 3.34c 3.16c 2.43b 101.70*** .22
Touch
aversion 1.80a 2.86c 2.17b 2.78c 74.42*** .17
Haven of
safety 4.67c 4.22b 5.00d 3.72a 66.95*** .06
Sexual preferences (df, 3, 1055)
Promiscuous 1.60a 2.24bc 2.08b 2.49c 21.37*** .06
Normative 5.61b 4.95a 5.60b 5.05a 15.42*** .04
Affectionate 6.25b 5.75a 6.11b 5.66a 25.17*** .07
Post-coital emotions (df 3, 698)
Positive 4.16c 3.08a 3.57b 3.29a 43.80*** .16
Negative .68a 1.45c 1.14b 1.18b 27.31*** .11
_____________________________________________________________________________________
Note. Numbers in the first four columns are means. Means within each row whose superscripts differ are
different at p < .05. *** p < .001 (two-tailed).
Table 8
Parameter Estimates for the Prediction of Touch and Sex Subscale Scores from Bartholomew's and Cluster-Based Categorical Attachment Measures, and Avoidance and Anxiety Attachment Dimensions
____________________________________________________________________________________
Bartholomew's Cluster-based Dimensional
measure (?2) measure (?2) measures (R2)
Touch
Affectionate touch .07 .23 .32
Desire for more touch .07 .23 .30
Touch aversion .05 .18 .20
Haven of safety .06 .16 .26
Sexual preferences
Promiscuous .03 .06 .06
Normative .01 .04 .06
Affectionate .04 .07 .10
Post-coital emotions
Positive .07 .16 .21
Negative .06 .11 .15
____________________________________________________________________________________
Sum (?): .46 1.24 1.66
Average (M): .05 .14 .18
____________________________________________________________________________________Note. Parameter estimates for the categorical measures are ?2s. Parameter estimates for the Avoidance and Anxiety dimensions are R2s. For the categorical variables, two-way analyses of variance were conducted in which self (positive vs. negative) and other (positive vs. negative) models were treated as separate factors. Thus, parameter estimates include variance due to both main effects and the interaction between Avoidance and Anxiety.
Figure 1. Ainsworth et al.'s (1978) Figure 10 (p. 102).
Figure 2. A Diagram of the Four Infant Attachment Types.
Figure 3. Bartholomew's Four-Category Diagram.
Appendix A
Multi-Item Measure of Adult Romantic Attachment
Experiences in Close Relationships
Instructions: The following statements concern how you feel in romantic relationships. We are interested in how you generally experience relationships, not just in what is happening in a current relationship. Respond to each statement by indicating how much you agree or disagree with it. Write the number in the space provided, using the following rating scale:
Disagree Neutral/Mixed Agree
Strongly Strongly
1 2 3 4 5 6 7
____ 1. I prefer not to show a partner how I feel deep down.
____ 2. I worry about being abandoned.
____ 3. I am very comfortable being close to romantic partners.
____ 4. I worry a lot about my relationships.
____ 5. Just when my partner starts to get close to me I find myself pulling away.
____ 6. I worry that romantic partners won't care about me as much as I care about them.
____ 7. I get uncomfortable when a romantic partner wants to be very close.
____ 8. I worry a fair amount about losing my partner.
____ 9. I don't feel comfortable opening up to romantic partners.
____ 10. I often wish that my partner's feelings for me were as strong as my feelings for him/her.
____ 11. I want to get close to my partner, but I keep pulling back.
____ 12. I often want to merge completely with romantic partners, and this sometimes scares them away.
____ 13. I am nervous when partners get too close to me.
____ 14. I worry about being alone.
____ 15. I feel comfortable sharing my private thoughts and feelings with my partner.
____ 16. My desire to be very close sometimes scares people away.
____ 17. I try to avoid getting too close to my partner.
____ 18. I need a lot of reassurance that I am loved by my partner.
____ 19. I find it relatively easy to get close to my partner.
____ 20. Sometimes I feel that I force my partners to show more feeling, more commitment.
____ 21. I find it difficult to allow myself to depend on romantic partners.
____ 22. I do not often worry about being abandoned.
____ 23. I prefer not to be too close to romantic partners.
____ 24. If I can't get my partner to show interest in me, I get upset or angry.
____ 25. I tell my partner just about everything.
____ 26. I find that my partner(s) don't want to get as close as I would like.
____ 27. I usually discuss my problems and concerns with my partner.
____ 28. When I'm not involved in a relationship, I feel somewhat anxious and insecure.
____ 29. I feel comfortable depending on romantic partners.
____ 30. I get frustrated when my partner is not around as much as I would like.
____ 31. I don't mind asking romantic partners for comfort, advice, or help.
____ 32. I get frustrated if romantic partners are not available when I need them.
____ 33. It helps to turn to my romantic partner in times of need.
____ 34. When romantic partners disapprove of me, I feel really bad about myself.
____ 35. I turn to my partner for many things, including comfort and reassurance.
____ 36. I resent it when my partner spends time away from me.
Appendix B
Attachment Scales and Scoring Instructions Two Higher-Order Attachment Dimension Scales
Avoidance (alpha = .94)
Item-Total
Item # Correlation Item
_____________________________________________________________________________________
. .73 I prefer not to show a partner how I feel deep down.
3. .71 I am very comfortable being close to romantic partners. (R)
5. .70 Just when my partner starts to get close to me I find myself pulling away.
7. .70 I get uncomfortable when a romantic partner wants to be very close.
9. .69 I don't feel comfortable opening up to romantic partners.
1. .68 I want to get close to my partner, but I keep pulling back.
3. .68 I am nervous when partners get too close to me.
5. .68 I feel comfortable sharing my private thoughts and feelings with my partner. (R)
7. .68 I try to avoid getting too close to my partner.
9. .67 I find it relatively easy to get close to my partner. (R)
21. .67 I find it difficult to allow myself to depend on romantic partners.
23. .65 I prefer not to be too close to romantic partners.
25. .64 I tell my partner just about everything. (R)
27. .64 I usually discuss my problems and concerns with my partner. (R)
29. .64 I feel comfortable depending on romantic partners. (R)
31. .63 I don't mind asking romantic partners for comfort, advice, or help. (R)
33. .62 It helps to turn to my romantic partner in times of need. (R)
35. .60 I turn to my partner for many things, including comfort and reassurance. (R)
Anxiety (alpha = .91)
Item-Total
Item # Correlation Item
_____________________________________________________________________________________
2. .67 I worry about being abandoned.
4. .65 I worry a lot about my relationships.
6. .65 I worry that romantic partners won't care about me as much as I care about
them.
8. .63 I worry a fair amount about losing my partner.
0. .62 I often wish that my partner's feelings for me were as strong as my feelings for him/her.
2. .60 I often want to merge completely with romantic partners, and this sometimes scares them away.
4. .60 I worry about being alone.
6. .57 My desire to be very close sometimes scares people away.
8. .56 I need a lot of reassurance that I am loved by my partner.
20. .55 Sometimes I feel that I force my partners to show more feeling, more
commitment.
22. .54 I do not often worry about being abandoned. (R)
24. .52 If I can't get my partner to show interest in me, I get upset or angry.
26. .52 I find that my partner(s) don't want to get as close as I would like.
28. .51 When I'm not involved in a relationship, I feel somewhat anxious and insecure.
30. .51 I get frustrated when my partner is not around as much as I would like.
32. .51 I get frustrated if romantic partners are not available when I need them.
34. .50 When romantic partners disapprove of me, I feel really bad about myself.
36. .50 I resent it when my partner spends time away from me.
Scoring Instructions
STEP 1: Recode the reversed variables, such that 1=7, 2=6, etc. You may want to create temporary variables, which can be reversed without potentially incorrectly transforming the original data. (We computed 'temp3' for item number 3, etc., for use in scoring below.)
Compute temp3 = A3.
Compute temp15 = A15.
Compute temp19 = A19.
Compute temp25 = A25.
Compute temp27 = A27.
Compute temp29 = A29.
Compute temp31 = A31.
Compute temp33 = A33.
Compute temp35 = A35.
Compute temp22= A22.
Recode temp3 to temp22 (1=7) (2=6) (3=5) (5=3) (6=2) (7=1).
STEP 2: Compute scores for the two dimensions, avoidance and anxiety.
Compute AVOIDANC = mean.14(A1,temp3,A5,A7,A9,A11,A13,temp15,A17,temp19,A21,
A23,temp25,temp27,temp29,temp31,temp33,temp35).
Compute ANXIETY = mean.14(A2,A4,A6,A8,A10,A12,A14,A16,A18,A20,temp22,A24,
A26,A28,A30,A32,A34,A36).
STEP 3: Compute attachment-style categories from the classification coefficients (Fischer's
linear discriminant functions) based on our sample of N = 1082.
Compute SEC2 = avoidanc*3.2893296 + anxiety*5.4725318 - 11.5307833.
Compute FEAR2 = avoidanc*7.2371075 + anxiety*8.1776446 - 32.3553266.
Compute PRE2 = avoidanc*3.9246754 + anxiety*9.7102446 - 28.4573220.
Compute DIS2 = avoidanc*7.3654621 + anxiety*4.9392039 - 22.2281088.
Variable Labels
sec2 'coeff secure dimension'
fear2 'coeff fearful dimension'
pre2 'coeff preoccupied dimension'
dis2 'coeff dismissing dimension'.
If (sec2 > max(fear2,pre2,dis2)) ATT2 = 1.
If (fear2 > max(sec2,pre2,dis2)) ATT2 = 2.
If (pre2 > max(sec2,fear2,dis2)) ATT2 = 3.
If (dis2 > max(sec2,fear2,pre2)) ATT2 = 4.
Variable labels
ATT2 'coefficient-based attachment category'.
Value labels
ATT2 1 'secure' 2 'fearful' 3 'preocc' 4 'dismiss'/.
61
Self-Report Measurement