|Home | About | Journals | Submit | Contact Us | Français|
Clinical theorists have suggested that disturbed attachments are central to borderline personality disorder (BPD) psychopathology. This article reviews 13 empirical studies that examine the types of attachment found in individuals with this disorder or with dimensional characteristics of BPD. Comparison among the 13 studies is handicapped by the variety of measures and attachment types that these studies have employed. Nevertheless, every study concludes that there is a strong association between BPD and insecure attachment. The types of attachment found to be most characteristic of BPD subjects are unresolved, preoccupied, and fearful. In each of these attachment types, individuals demonstrate a longing for intimacy and—at the same time—concern about dependency and rejection. The high prevalence and severity of insecure attachments found in these adult samples support the central role of disturbed interpersonal relationships in clinical theories of BPD. This review concludes that these types of insecure attachment may represent phenotypic markers of vulnerability to BPD, suggesting several directions for future research.
Ever since the inception of the borderline personality disorder (BPD) diagnosis, clinical theorists1-5 have suggested that the disorder's core psychopathology arises within the domain of interpersonal relations. These theories were prompted by the centrality of interpersonal demands and fears within clinical contexts. While there has been growing evidence and interest in biogenetic bases for borderline pathology,6,7 these perspectives do not diminish the potential role that disturbed relationships have as risk markers or as mediating factors in BPD's pathogenesis.
In recent years the methodology for reliably measuring attachment styles has provided a welcome opportunity to characterize empirically the interpersonal problems of BPD patients. Because the insecure attachments of borderline patients are so manifest, so central to the problems that they present for treatment, and so central to theories about the pathogenesis of BPD, the empirical examination of these attachments has considerable clinical and theoretical significance. The resulting literature—still growing rapidly—is the subject of this review.
In the background of the attention being given to attachment problems in borderline patients is the seminal developmental theory of John Bowlby.8-10 He postulated that human beings, like all primates, are under pressures of natural selection to evolve behavioral patterns, such as proximity seeking, smiling, and clinging, that evoke caretaking behavior in adults, such as touching, holding, and soothing. These reciprocal behaviors promote the development of an enduring, affective tie between infant and caregiver, which constitutes attachment. Moreover, from these parental responses, the infant develops internal models of the self and others that function as templates for later relationships.9 These models, which tend to persist over the life span, guide expectations or beliefs regarding interactions in past, present, and future relationships. For Bowlby,9 the content of the internal working model of self is related to how acceptable or lovable one is in the eyes of primary attachment figures. The content of an individual's model of other is related to how responsive and available attachment figures are expected to be.
The goal of attachment is the creation of an external environment from which the child develops an internal model of the self that is safe and secure. Secure attachment to the caregiver liberates the child to explore his or her world with the confidence that the caregiver is available when needed. A secure attachment should engender a positive, coherent, and consistent self-image and a sense of being worthy of love, combined with a positive expectation that significant others will be generally accepting and responsive. This portrait of secure attachment contrasts dramatically with the malevolent or split representations of self and others,11 as well as with the needy, manipulative, and angry relationships, that characterize persons with BPD.1,2,5
Fonagy and colleagues12-14 have proposed that a child is more likely to develop a secure attachment if his or her caregivers have a well-developed capacity to think about the contents of their own minds and those of others. This secure attachment, in turn, promotes the child's own mental capacity to consider what is in the mind of his or her caregivers. In contrast, individuals with BPD demonstrate a diminished capacity to form representations of their caretakers' inner thoughts and feelings. In this way a child defensively protects himself or herself from having to recognize the hostility toward, or wish to harm, him or her that may be present in the parent's mind. In Fonagy's theory this diminished capacity to have mental representations of the feelings and thoughts of self and others accounts for many of the core symptoms of BPD, including an unstable sense of self, impulsivity, and chronic feelings of emptiness.
Several clinical theorists have posited intolerance of aloneness as a defining characteristic for BPD that provides coherence to the DSM's descriptive criteria.2,15 Gunderson3 subsequently suggested that this intolerance reflects early attachment failures, noting that individuals with BPD are unable to invoke a “soothing introject” in times of distress because of inconsistent and unstable attachments to early caregivers or, in Bowlby's terms, because of insecure attachment. Gunderson observed that descriptions of certain insecure patterns of attachment—specifically, pleas for attention and help, clinging, and checking for proximity that often alternate with a denial of, and fearfulness about, dependency needs—closely parallel the behavior of borderline patients.
Comparing theories of object relations and attachment, Lyons-Ruth16,17 has distinguished normal processes of separation-individuation in early development from the disorganized conflict behaviors displayed toward attachment figures by toddlers at risk for later psychopathology. She has argued that disorganized insecure attachment in infancy (see below) represents a deviant developmental pattern that, when present, may be an identifiable risk factor for the later development of BPD.
The empirical assessment of patterns of attachment behaviors began with Ainsworth and colleagues'18 typology of infant attachment behaviors toward their mothers when under stress. Under this typology, there were three organizations of infant attachment behavior: secure, avoidant, and ambivalent attachment (Table 1). In subsequent years, these infant behavioral patterns have been intensively researched, and a core body of empirical findings has been extensively replicated.21
As infant attachment assessments were extended to high-risk or psychiatric samples, many of the infant behavioral patterns observed did not conform to any of the three attachment patterns characteristic of infants in low-risk settings. These repeated observations led Main and Solomon19 to review a large number of at-risk infant videotapes and develop coding criteria for a fourth category labeled disorganized/disoriented (Table 1). Disorganized attachment behaviors were subsequently found to be associated with family environments characterized by increased parental risk factors such as maternal depression, marital conflict, or child maltreatment. These attachment behaviors are also the behaviors most consistently associated with childhood psychopathology, including internalizing and externalizing symptoms at school age, as well as overall psychopathology and dissociative symptoms by late adolescence.17
A major step in the developmental research literature on attachment occurred with the introduction by Main and colleagues22 of the Adult Attachment Interview (AAI) in 1985. The AAI is a semistructured interview developed to assess the adult counterparts of the secure, avoidant, and ambivalent attachment strategies observed during infancy and childhood. The interview lasts approximately one hour and poses a series of questions probing how the individual thinks about his or her childhood relationships with parents or other central attachment figures. The interview is coded not for the positive or negative content of childhood experiences or memories, but in terms of narrative analysis—that is, for how the individual organizes his or her attention and discourse regarding attachment topics over the course of the interview.
Adult strategies for discussing positive and negative attachment experiences in childhood are observable in the interview and parallel the infant strategies described earlier. Flexible and coherent discourse around both positive and negative attachment experiences is termed autonomous (the equivalent of secure in childhood); deactivating strategies are termed dismissing (the equivalent of avoidant); and hyperactivating strategies are termed preoccupied (the equivalent of ambivalent).
Shortly after the introduction of the AAI, Ainsworth and Eichberg23 reported that the parents' lapses in the monitoring of discourse or reasoning during discussions of loss or trauma on the AAI predicted disorganized attachment behaviors in their infants. This finding has now been well replicated, leading Main and Goldwyn24 to develop a fourth category for the AAI labeled unresolved with respect to loss or trauma. Unresolved attachment patterns are the only patterns that are also given a secondary subclassification (namely, unresolved/autonomous, unresolved/dismissing, or unresolved/preoccupied) that indicates which organized attachment classification is the best-fitting alternative classification. That is, since an unresolved classification is understood as indicating a collapse of strategy—as seen in the failure to use a single, consistent strategy over the course of the interview—the secondary classification is used to indicate the best guess as to the strategy that has failed.
Although Bowlby was primarily interested in young children, he maintained, as noted earlier, that the core functions of the attachment system continue throughout one's life span.9 In a series of independent developments in the field of social psychology, Hazan and Shaver25 were first to apply concepts of attachment developed from studies of the parent-child relationship to the romantic relationships found between adults. For example, feeling securely attached arises after receiving feedback from other adults that one is loved and capable.26 This inner sense of security contributes to a stable, consistent, and coherent self-image and to the ability to reflect upon and correctly interpret others. The social psychological tradition has defined secure, dismissing/avoidant, anxious/preoccupied, and fearful/avoidant attachment (Table 1).8,20 To simplify, these types will hereafter be referred to as dismissing, preoccupied, and fearful.
Each of the self-report measures has its own distinguishing features that, while beyond the scope of this review, are described in a 1995 article by Crowell and Treboux.27 In what follows, we focus on the measures most relevant for our purposes. As noted above, Hazan and Shaver25 applied the three original patterns of attachment to the study of romantic relationships between adults, opening up a major paradigm of research focusing on adult attachment. The self-report instrument that was used, the Attachment Self-Report (ASR), asked subjects to pick the one of three paragraphs (representing secure, anxious/ambivalent, and avoidant) that best represented their relationships. Bartholomew and colleagues28,29 took a step toward integrating the social-psychological and developmental attachment work by proposing a two-dimensional construct of adult attachment—one based on the intersection of a model of the self and a model of others. Security was defined as a positive model of self and a positive model of others. Anxious/ambivalent was relabeled as preoccupied and defined as representing a negative model of self, combined with a positive model of others. The avoidant classification was divided into two groups: fearful, representing a negative model of self with a negative model of others, and dismissing, representing a positive model of self with a negative model of others. Two popular measures were constructed to fit with this line of research. The Relationship Questionnaire28 (RQ) asks participants to rate (on a scale of 1 to 7) how much they endorse four different paragraphs, each representing one of the four styles. The Relationship Scales Questionnaire29 (RSQ) uses 17 items concerning feelings, thoughts, and behaviors in relationships to capture the dimensions of the internal working models (model of self and model of other) that are latent in each subject's particular style.
Simultaneously, other social psychologists developed additional self-report measures for assessing adult attachment. Of relevance to the research reviewed in this article is the Attachment Style Questionnaire.30 This multi-item, self-report questionnaire, a derivative of the ASR and RQ, scores five dimensions (confidence, discomfort with closeness, need for approval, preoccupation with relationships, and relationships as secondary) that capture the behaviors and feelings latent in attachment styles.
Another development within the social psychological perspective has been the movement toward using dimensional scoring, rather than prototype measures, of attachment types. Hence, some studies reviewed in this article use a dimensional, rather than a prototypic, approach to attachment, asking “how much” of the secure, dismissing, preoccupied, and fearful attachment styles exist within the same individual, rather than strictly classifying each person as belonging to one or another style.
It is important to note that the attachment types derived from self-report measures or developed by social psychologists differ in several ways from the types derived from the AAI originated by developmental researchers. As noted above, the AAI is scored by analysis of an individual's narrative account rather than by the content of his or her statements regarding attachment to parents in the past. In contrast, the self-report measures rely on conscious perceptions of one's attachment (either retrospectively with parents or in current peer and romantic relationships) and thus are subject to response bias.14 For example, a frightened person is apt to assign fearful qualities to his or her relationships. Moreover, the self-report measures provide information on the attachment-related style associated with a particular relationship rather than suggesting a single, underlying representational model for all attachment relationships derived from the early parent-child relationship, which is how the types derived from the AAI are interpreted.
Though both developmental and social psychological measures have similar theoretical roots in Bowlby's work, it is important to note that the aspects of attachment assessed by each tradition are different, and that the two sets of measures are not closely correlated with one another. On the plus side, Bartholomew and Horowitz28 found very good correspondence between AAI and RQ measures of preoccupied and dismissing types. On the minus side, however, Waters and colleagues31 found quite different correlates of the AAI and the Experiences in Close Relationships (ECL) self-report questionnaire.32 In particular, Waters and colleagues31 found that the AAI, consistent with expectations, correlated well with measures of parent-child interaction—that is, with laboratory observations of attachment security in infancy, with laboratory observations of the toddler's use of secure base support from the parent, and with the parent's knowledge of secure base scripts. In contrast, the ECL, consistent with expectations, correlated strongly with measures of adult marital satisfaction and dissatisfaction, depression, commitment, and passion and intimacy.
Table 2 summarizes (including sample size, comparison groups, and assessment tools) the 13 empirical studies that have linked BPD with attachment classifications. We will comment on the methodological and design issues found in the existing studies, and then examine how these studies characterize the types of attachment found in BPD samples.
We used MEDLINE for journals published in English with the search items “borderline personality” and “attachment.” We identified additional studies in the reference sections of these articles. The 13 studies that were thus identified are the basis for this review. Because the measures used to assess attachment differed substantially from study to study in their theoretical origins, descriptive terminology, procedures by which data were collected, and the particular relationships in which attachment was rated, we will consider the ways in which these differences influence the interpretation of the studies. We will also identify the ways in which differences in the samples of subjects affect the results. With due consideration for these methodological problems, we then describe the studies' results concerning the attachment patterns that characterize borderline patients.
It is noteworthy that the sample size for most of these studies is either quite small or unclear due to reliance on a dimensional scheme. In eight of the nine studies that report the number of BPD subjects, that number ranges from 837 to 49.39 The remaining, ninth study40 has 426 BPD subjects, but this sample (representing 30.5% of a college population) is grossly overinclusive; in the general population, the estimated prevalence for BPD is 0.6–3%.45-47 The four studies that do not provide sample sizes of BPD subjects describe BPD dimensionally; that is, subjects are rated as being borderline to a greater or lesser extent. In these studies, the overall samples are larger, ranging from 6038 to 393.43
Four of the studies drew the BPD subjects from both inpatient and outpatient psychiatric settings;33,37,41,42 three from inpatient psychiatric settings alone;13,38,39 three from outpatient psychiatric settings alone;34,36,44 two from university students;40,43 and one from court-referred abusive men.35 The possible significance of sample selection is demonstrated by two studies, both with carefully diagnosed BPD samples that used the same attachment measure (the ASR). Ambivalent attachment discriminated those with BPD traits among university students,43 and avoidant attachment discriminated BPD patients who were selected from inpatients.39
Only one study36 had a homogenous diagnostic comparison group—namely, dysthymic disorder. All others used a mixed population of other psychiatric disorders13,37,38,43 or normals.13,39-41 Two studies used comparison groups with a variety of other personality disorders or traits.40-42 Only two studies used comparison groups that were matched with the BPD samples. Patrick and colleagues36 matched the two groups for age and educational achievement. Fonagy and colleagues13 matched the BPD group and normal control group for age, gender, social class, and verbal IQ, although they did not match the non-BPD psychiatric control group.
The relationship targeted in the AAI studies is that between subjects and their parents. In the six studies based on self-reports, three are directed at peers,35,40,43 two are directed at all (that is, unspecified) relationships,41,42 and one includes separate assessments for peer, parental, and all relationships, each with a distinct instrument.39 The significance of the target relationships is illustrated by the study by Sack and colleagues.39 They concluded that relationships with mothers were most often classified as ambivalent (41%), with only 18% considered avoidant, whereas attachment to their fathers was most often classified as avoidant (44%), with only 18% considered ambivalent. By so clearly distinguishing the attachment to mother and father, this study shows that variations in the types of insecure attachment shown by BPD subjects may be partly accounted for by choice of the target relationship.
Table 3 identifies attachment types that have been found to distinguish BPD from non-BPD samples in the 13 studies. Each type is accompanied by an abbreviated description.
Since all the theories discussed earlier, as well as the standard DSM description, indicate that, by definition, borderline subjects' relationships are not secure, it is of some interest that a fraction of borderline patients in these studies were found to be categorized as secure. Although two of the five studies utilizing the AAI showed that none of the individuals with BPD had secure attachment,36,37 the other three of those studies13,38,44 showed small percentages—either 7% or 8%—that did. Moreover, two studies using self-report measures39,40 found that 9% and 29.8% of the BPD subjects had secure attachment. The other four studies did not report the proportion of secure attachment among the BPD patients. All studies demonstrated an inverse relationship between secure attachment and BPD when the disorder was rated in a dimensional fashion. Fossati and colleagues41 reported a lower mean confident (that is, secure) score among BPD subjects than nonpatients (p = .0025). Dutton,35 Nickell,43 and their colleagues showed that their dimensional ratings of borderline pathology were highly negatively correlated to secure attachment (p = .001 and p = 0.01, respectively). Meyer and colleagues42 demonstrated a negative correlation between secure attachment and each of the 13 personality disorders that they examined; the negative correlation was most robust for the borderline scale (p = .01).
All of the studies revealed an association between the diagnosis of BPD and insecure forms of attachment. Of the seven studies employing the categories preoccupied or unresolved, the five using the AAI all showed that the greatest proportion of borderline individuals fall into these attachment types.13,34,35,38,44 In the two studies using self-report measures of preoccupied attachments35,40—which, as shown in Table 1, is a somewhat different construct—the results were different. For Patrick and colleagues,36 all 9 of the borderline patients who had experienced loss or trauma were given a primary classification as unresolved with respect to loss or abuse, as well as a secondary classification as preoccupied. Three additional patients with BPD were given a primary classification of preoccupied. Ten out of the 12 patients with any preoccupied classification were assigned to a rare preoccupied subtype termed “confused, fearful, and overwhelmed” by traumatic experiences. Stalker and colleagues37 found 7 out of the 8 women with BPD were given a primary classification of unresolved, and 5 of 8 were given a primary or secondary classification of preoccupied. Fonagy and colleagues13 described 32 of 36 patients with BPD (89%) as having a primary classification of unresolved, and 27 of 36 patients (75%) as having a primary or secondary classification of preoccupied. Barone44 found that out of 40 BPD patients, 50% were given a primary classification of unresolved; 23%, of preoccupied; and 20%, of dismissing. Rosenstein and Horowitz38 found 8 of 14 adolescents with BPD (64%) to have a preoccupied attachment style. This study did not assess unresolved attachment. The two studies that used self-report measures found that fearful attachment characterized BPD. For Dutton and colleagues,35 both fearful and preoccupied attachment, as assessed by the RQ and RSQ in abusive men, were predictive for borderline personality, but fearful attachment was so strong a predictor that the authors concluded that having borderline personality was the prototype for this particular attachment style. Using the RQ and their overinclusive sample of students, Brennan and Shaver40 found that 32.2% were fearful; 24.6%, preoccupied; 13.4%, dismissing; and 29.8%, secure.
Fossati and colleagues41 found that inpatients and outpatients with BPD scored significantly higher than non-patients on all insecure dimensions—that is, preoccupation (p = .0025), discomfort with closeness (p = .0025), need for approval (p = .0025), and relationships as secondary (p = .0025). This result suggests that the combination of unresolved and preoccupied or fearful classifications may serve to identify a complex combination of insecure features. Consistent with the complexity of insecure features in the study by Fossati and colleagues,41 West and colleagues34 found that high scores on each of four attachment scales—feared loss, secure base (coded negatively), compulsive caregiving, and angry withdrawal—successfully distinguished patients with BPD among 85 female outpatients. Among the studies that did not include categories or scales for fearful or unresolved attachment, Sperling and colleagues33 used a three-category coding of the AAI among 24 hospitalized BPD patients. They found that a dependent style of attachment was associated with less BPD pathology than an avoidant or an ambivalent style. Finally, Meyer and colleagues42 found that three patients with BPD scored very highly on the study's measure of borderline attachment prototype, which is defined as “ambivalent and erratic feelings in close relationships.”
These studies of borderline personality employ a variety of measures and types of insecure attachment. Moreover, the target relationship varies in the different studies from one with peers, parents, or a generic other. These variations make comparisons between studies difficult (see reviews by Stein and colleagues14 and by Crowell & Treboux).27 The attachment field sorely needs studies that document the correlations among the different attachment types identified by the various instruments. The particular area reviewed here also still needs large samples of carefully diagnosed borderline patients with matched comparison groups. For the present review, we must rely on our hypothesized correlations among the attachment types—hypotheses based on the concordance of, or differences between, the definitions posited by each instrument. Moreover, the studies under review have utilized varied sources for sample acquisition (colleges versus hospitals, for example), various comparison groups and diagnostic methods, and generally small sample sizes. Finally, these studies have used measures developed to describe attachment styles among nonclinical populations. Arguably, however, rather than attempting to fit attachment patterns seen in high-risk or clinical samples into descriptors developed for normative populations, what is needed is further description of the specific attachment behaviors and internal models characteristic of the clinical groups themselves; these patterns are likely to be more complex and contradictory than those prevalent in nonclinical samples (for example, see additional AAI codes for hostile-helpless states of mind developed by Lyons-Ruth and colleagues).50 The conclusions to be drawn from this review are thereby greatly limited and should be considered, at best, as informed hypotheses.
Despite the great variation in study design and methodology, all 13 of the studies relating attachment to BPD concluded that there was a strong association with insecure forms of attachment. This finding is consonant with theories that see interpersonal instability as the core of BPD psychopathology. Still, given that BPD samples were defined, in part, by DSM criteria that include intense and unstable relationships as a diagnostic feature, this result is somewhat circular. A recent report by Meyer and colleagues42 illustrates this point. They found that their Borderline Attachment Prototype correlated so highly with borderline criteria that only a single variable could be used in a regression analysis. Nonetheless, this result suggests that despite measures that differ substantially, all are capturing some essential subsyndromal—that is, phenotypic—problems in the interpersonal relationships of borderline individuals. The one exception to this pattern of insecure attachments—the study by Brennan and Shaver,40 with nearly 30% of the subjects having secure attachment—is likely a consequence of the study's highly overinclusive method of sampling. Indeed, given the emphasis on interpersonal problems in borderline psychopathology, it would seem that anytime secure attachment is found, either the diagnosis or the attachment measure should be considered suspect.
The most consistent findings from this review are that borderline patients have unresolved and fearful types of attachment. In all studies using the AAI, from 50% to 80% of borderline patients were classified as unresolved. In the two studies using self-report instruments that assessed fearful attachment, that classification was the one most frequently associated with borderline features (among abusing men and college students).
It is notable that all unresolved subjects were also secondarily classified as preoccupied. Moreover, in the self-report studies that included a fearful classification, preoccupied attachment was the second most strongly endorsed category among borderline subjects. In no study that included the unresolved or fearful classification, however, was preoccupied the most prevalent classification. Preoccupied (or ambivalent) attachments are defined as ones in which individuals seek close, intimate relationships but are very reactive to their perceived dependency or undervaluation. This description is close to what Meyer and colleagues42 defined as the prototypic borderline form of attachment—that is, “ambivalent and erratic feelings in close relationships.” The characterization as fearful also entails a longing for intimacy, but fearful individuals are concerned about rejection rather than excessive dependence. Patrick and colleagues36 bridged these types by demonstrating that borderline patients had a fearful subtype of preoccupied attachment (as well as being unresolved). In sum, then, BPD attachments seem best characterized as unresolved with preoccupied features in relation to their parents, and fearful or, secondarily, preoccupied in their romantic relationships. While in our view and that of others,51 the self-report fearful category and the AAI unresolved category seem to overlap, such an overlap remains to be demonstrated empirically.
The high prevalence and severity of unresolved/preoccupied (AAI) or fearful (self-report) attachments found in these adult samples support the central role that interpersonal relationships have had in clinical theories on BPD. Insecure attachments, especially of unresolved or fearful type—or their disorganized analogues in infancy and childhood—might serve as markers of risk for development of BPD. This hypothesis invites other research in which these forms of insecure attachment in adults could be used as a subsyndromal phenotype signifying a predisposition to BPD that takes its place alongside the phenotypes of affective instability and impulsivity as predisposing toward BPD.6 Such possibilities are confirmed by evidence that disturbed attachments may have heritable components.52-54 Family-study methodology could usefully test whether a BPD-related risk factor exists in the form of unresolved or fearful attachments that are transmitted across generational boundaries.