Self-Report Measurement of Adult Attachment: An Integrative Overview.

Authors Avatar
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.
Join now!


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