The parenting tradition

The parenting tradition originated from research into how adult attachment might influence parenting behaviour and the attachment patterns of parents’ young children (Reference Bartholomew, Shaver, Simpson and RholesBartholomew & Shaver, 1998). The most researched measure here is the Adult Attachment Interview (Adult Attachment Interview; Reference Main and GoldwynMain & Goldwyn, 1998), which measures the adult’s state of mind with respect to attachment, i.e. their current representations of their childhood relationships with their parents. The instrument is a semi-structured interview about early attachment history, with questions designed to ‘surprise the unconscious’ and hence yield clues as to the predominant underlying attachment strategies of an individual that influence current attachment relationships. Interviewees are asked about past attachment experiences and their current influence, and to give specific examples to support their statements. The interview yields both categorical and continuous data. In the Main & Goldwyn scoring system, the rating procedure is as dependent on how interviewees answer the questions (the discourse style) as on what they actually say (the content) (Reference Hesse, Cassidy and ShaverHesse, 1999). Interviewees are put into one of four broad adult attachment categories, which are analogous to the infant categories produced in Ainsworth’s Strange Situation procedure (Reference Ainsworth, Blehar and WatersAinsworth et al, 1978) (Table 2). Indeed, the Adult Attachment Interview draws its validity from the high correspondence between the attachment classification of parents and their infants’ classifications in the Strange Situation (Reference van IJzendoornvan IJzendoorn, 1995). Thus, ‘secure’, ‘ambivalent’, ‘avoidant’ and ‘disorganised’ infants tended to have primary caregivers who were respectively ‘autonomous’, ‘preoccupied’, ‘dismissing’ and ‘unresolved’ with respect to attachment. Many of the leading researchers in the parenting attachment research tradition were students of Mary Ainsworth.

Table 2

Summary of attachment categories¹

Infant attachment category	Observations of infant in the Strange Situation 2	Corresponding adult attachment category	Discourse style on the Adult Attachment Interview
Avoidant (A)	Little protest on separation. On caregiver’s return, hovers warily nearby, cannot play freely	Dismissing (Ds)	Brief discourse, which normally idealises parents, with few supporting examples
Secure (B)	Protests when caregiver disappears. Protest continues on return, but soon pacified and continues exploratory play	Secure/autonomous (F)	Narrative coherence. Valuing of attachment, but seems objective regarding particular events and relationships. Able to give examples to support statements
Ambivalent/resistant (C)	Protests, and hard to pacify on caregiver’s return, clings to carer, buries head in lap, pushes away toys offered	Preoccupied (E)	Incoherent, vague and excessively long discourse. Preoccupied with past attachment experiences. Speaker appears angry, passive or fearful
Disorganised/disoriented (D)	‘Freezes’ on separation, seems unable to sustain any organised pattern of behaviour on reunion. Behaviours may appear bizarre and stereotyped	Unresolved/disorganised (U)3	Striking lapses in monitoring of reasoning during discussion of loss or abuse

In infants, the avoidant and ambivalent patterns are said to be organised, as even these ‘insecure’ patterns may be adaptive in maximising care from broadly rejecting and inconsistent caregivers respectively. Disorganised attachment, however, may represent a more fundamental breakdown in attachment strategy. Unresolved loss and trauma in the caregiver predicts infant attachment disorganisation. When discussing past traumas or losses on the Adult Attachment Interview, such a caregiver will show striking lapses of ‘meta-cognitive reasoning’ and possible dissociation; she will accordingly receive an unresolved/disorganised classification. Disorganisation of infant attachment, more than ‘organised’ insecure attachment patterns (i.e. avoidant or ambivalent), is predictive of both externalising and internalising problems in childhood (Reference Lyons-Ruth, Jacobvitz, Cassidy and ShaverLyons-Ruth & Jacobvitz, 1999; Reference Green and GoldwynGreen & Goldwyn, 2002). Similarly, the unresolved attachment category in adults may be especially associated with psychopathology.

The romantic attachment tradition

In the past two decades, an independent line of attachment measurement has originated from research into romantic relationships (Reference Bartholomew, Shaver, Simpson and RholesBartholomew & Shaver, 1998). This was based on the premises that romantic love could be conceptualised as an attachment process (Reference Hazan and ShaverHazan & Shaver, 1987) and that how one views and approaches romantic relationships might be an outgrowth of previous attachment experiences. Many in the romantic attachment tradition have been personality or social psychologists (Table 1). This strand of attachment research, which focused as a starting point on adulthood, is characterised by the use of self-report measures of attachment, although narrative interview can also be used.

The Adult Attachment Interview and self-report questionnaires

To score the Adult Attachment Interview requires an extensive training that is not easily available to the average clinician. Self-report attachment measures, on the other hand, are more easily scored and analysed, which makes them easier to use in larger-scale research. They may also be useful in the clinical setting, whereas the Adult Attachment Interview, because of its complexity, has remained primarily a research tool.

The use of self-report questionnaires has been challenged on the basis of their (theoretically) limited ability to tap into unconscious attachment strategies and their vulnerability to defensive reporting. The Adult Attachment Interview, by its design, would not suffer from these deficiencies. However, in support of self-report measures, it can be counter-argued that adults are able to provide valuable information about their emotional experiences, that most adults have sufficient experience in close relationships to recount how they behave in them, and that conscious and unconscious processes normally operate together to achieve a goal. The construct validity of self-report measures of romantic attachment derives from their prediction of three domains: differential behaviours within relationships consistent with attachment theory (e.g. the use of a partner as a secure base); the differing attributions made by secure and insecure individuals within relational conflicts; and general adjustment and, as we shall see, psychopathology (Reference Crowell, Fraley, Shaver, Cassidy and ShaverCrowell et al, 1999). At the same time, self-report measures of romantic attachment show only moderate correlation with measures of personality and relationship satisfaction, suggesting sufficient discriminant validity (Reference Crowell, Fraley, Shaver, Cassidy and ShaverCrowell et al, 1999).

Bartholomew’s four-category model

One particular model in the romantic tradition that has been well validated and used in research into adult psychopathology is Bartholomew’s four-category model of attachment (Reference Bartholomew and HorowitzBartholomew & Horowitz, 1991). Figure 1 shows that this model is in fact conceptualised as different combinations of the internal working models of the self and of others, yielding the four attachment ‘prototypes’, or styles, of ‘secure’, ‘dismissing’, ‘preoccupied’ and ‘fearful’. These four attachment styles can be assessed using both interview methods and self-report questionnaires such as the Relationship Questionnaire (Reference Bartholomew and HorowitzBartholomew & Horowitz, 1991) and the Relationship Scales Questionnaire (Box 1) (Reference Griffin, Bartholomew, Bartholomew and PerlmanGriffin & Bartholomew, 1994).

Fig. 1

Bartholomew’s four-category model of adult attachment (after Reference Bartholomew and HorowitzBartholomew & Horowitz, 1991).

Many other self-report measures, devised by different research groups, now exist. A factor analysis of self-report measures (Reference Brennan, Clark, Shaver, Simpson and RholesBrennan et al, 1998) identified a latent two-factor structure underlying them. The two factors, attachment anxiety (i.e. anxiety over abandonment within attachment relationships) and attachment avoidance (avoidance of attachment relationships), are postulated to be affective–behavioural manifestations of the internal working models of the self and of others respectively.

There is doubt as to how far interview and self-report measures converge (Reference Crowell, Fraley, Shaver, Cassidy and ShaverCrowell et al, 1999). Nevertheless, Reference Bartholomew, Shaver, Simpson and RholesBartholomew & Shaver (1998) argue that common ground does exist. They believe that there is some convergence among the secure, dismissing and preoccupied categories in the Main & Goldwyn (Table 2) and Bartholomew classifications (Fig. 1). The fearful prototype, which may be especially relevant to adult psychopathology, is unique to the Bartholomew system, which does not contain attachment disorganisation. It would be tempting to equate fearful attachment with attachment disorganisation, and some researchers have in fact done so. However, this practice has been challenged by, among others, Reference West and GeorgeWest & George (2002).

Depressive disorder

Attachment theory may be highly relevant in the understanding of the aetiology of depression, as Reference BowlbyBowlby (1980) himself discussed. The experiences of early loss, separation and rejection by the parent or caregiver (conveying the message that the child is unlovable) may all lead to insecure internal working models (Reference Dozier, Stovall, Albus, Cassidy and ShaverDozier et al, 1999). Internal cognitive representations of the self as unlovable and of attachment figures as unloving/untrustworthy would be consistent with parts of Beck’s cognitive triad in depression (Reference Beck, Rush and ShawBeck et al, 1979).

Empirically, the association between attachment and depression has been assessed cross-sectionally using both the Adult Attachment Interview and self-report instruments, affording different facets of aetiological understanding. Studies using the Adult Attachment Interview, which have examined both adults and adolescents, have yielded inconsistent results (Reference Dozier, Stovall, Albus, Cassidy and ShaverDozier et al, 1999). However, small sample sizes and different diagnostic, inclusion and exclusion criteria have probably contributed to the contradictory findings. For example, the study by Reference Cole-Detke and KobakCole-Detke & Kobak (1996) looked at depressive symptoms rather than disorder, and found that college women who reported depressive symptoms were more often hyperactivating in their attachment strategies than were controls with no eating disorder/depressive symptoms or a sample with only symptoms of eating disorder. This study used Kobak’s Q-set method of scoring (further details available on request) rather than the Main & Goldwin (1998) system. The two systems do not give directly comparable results; disorganisation is not measured by the Q-set method, thus affecting the interpretation of results.

The comparatively large and much quoted study by Reference Fonagy, Leigh and SteeleFonagy et al(1996) did operationalise diagnoses according to the DSM–III–R (American Psychiatric Association, 1987). They assessed the relation between attachment and psychopathology using the Adult Attachment Interview in 82 non-psychotic psychiatric in-patients and 85 case-matched non-psychiatric controls. More than 80% of both groups were women. Applying the Main & Goldwyn scoring system, the authors analysed results using the three-way and four-way classificatory systems for the Adult Attachment Interview (excluding and including unresolved attachment respectively). Unfortunately, all affective disorders (including major depressive disorder, bipolar disorder and dysthymia) were analysed together, and comorbidity was significant among the psychiatric patients. The findings were therefore hard to interpret. They did identify a large proportion of patients who were in the ‘unresolved/disorganised’ U category on the Adult Attachment Interview (Table 2); this result could be attributed to the sample, which comprised in-patients referred to a national centre for the treatment of personality disorder (see below).

Studies using Bartholomew’s classification (Fig. 1) and self-report attachment measures have produced more consistent results. In non-clinical populations, where research has benefited from larger samples, depressive symptoms are positively associated with both self-reported preoccupied and fearful attachment, and negatively associated with secure attachment. Depressive disorder in clinical populations tends to be more associated with fearful attachment (e.g. Reference Carnelley, Pietromonaco and JaffeCarnelley et al, 1994; Reference Reis and GrenyerReis & Grenyer, 2004). In a study of 71 individuals with DSM–IV major depression, Reference Reis and GrenyerReis & Grenyer (2004) found an association between fearful attachment and depression severity, at least for the women. They note that whereas a negative internal working model of the self (seen in both preoccupied and fearful attachment) may be associated with milder, subclinical depressive symptoms, ‘the experience of major depression may be more intense or severe if others are simultaneously viewed as hostile, uncaring or rejecting’, as would be the case in fearful attachment.

Anxiety

Anxiety disorders are heterogeneous in nature, characterised by a combination of fear and avoidance. Reference Dozier, Stovall, Albus, Cassidy and ShaverDozier et al(1999) suggest that when fear predominates, the disorder involves primarily internalising symptoms, and will be expected to be associated with maximising attachment strategies (e.g. ambivalent/preoccupied attachment). Reference BowlbyBowlby (1973) proposed that anxiety disorders are best accounted for by anxiety regarding the availability of attachment figures, and delineated early family environments that might predispose to their development. These include environments in which the child worries about the parent’s safety in the child’s absence, or worries about parental rejection. An insecurely attached child may thus frequently become anxious, even in benign circumstances. Chronic vigilance and anxiety will then increase the probability of a future anxiety disorder (Reference Warren, Huston and EgelandWarren et al, 1997).

Given the theoretical links between attachment insecurity and anxiety disorders, there has been surprisingly little empirical research examining the association (Reference GoodwinGoodwin, 2003). However, Reference Warren, Huston and EgelandWarren et al(1997) conducted one of the few studies on attachment and psychopathology that benefits from a longitudinal design. They recruited 267 mother–baby dyads and assessed each infant’s attachment using the Strange Situation procedure. Current and past anxiety disorders were assessed in 172 of these children when they reached 17.5 years. Twenty-six (15%) had at least one past or current anxiety disorder, including separation anxiety and social phobia; 51% had a disorder other than anxiety disorder. Results showed the contribution of anxious/ambivalent attachment to anxiety disorders, which conferred a twofold increase in risk. However, the study did not consider attachment disorganisation as a risk factor, nor did it attempt to assess the adolescents’ attachment style at follow-up.

In the aforementioned study by Reference Fonagy, Leigh and SteeleFonagy et al(1996), 66% of the patients with anxiety disorder were classified as preoccupied (E) in the three-way Adult Attachment Interview classification. When unresolved/disorganised attachment (U) was taken into account, there was a significant association between the U category and anxiety disorders, with 86% of individuals receiving this attachment classification.

Neither of the above studies examined specific anxiety diagnoses. More recently, Reference Myhr, Sookman and PinardMyhr et al(2004) compared three groups (individuals with obsessive–compulsive disorder (OCD), with depression and with no psychiatric disorder) on a self-report measure of attachment, the Revised Adult Attachment Scale (Reference Collins and ReadCollins & Read, 1990), which differs from the Bartholomew measures. Both the OCD and the depression group scored significantly higher on the anxiety sub-scale of this attachment measure, which suggests an insecure self-model. Interestingly, the OCD group did not differ from the control group on recollection of parental bonding; the depression group, however, had more negative recollections. Parental bonding has been used by some researchers as a proxy measure of early attachment; such practice is of questionable validity (Reference Manassis, Owens and AdamManassis et al, 1999).

Stress-related disorders

Research has begun to address the relationship between post-traumatic stress symptoms and attachment. It is hypothesised that attachment insecurity would compromise an individual’s ability to cope with traumatic life events and would therefore predispose to the formation of post-traumatic stress symptoms and/or disorder. However, longitudinal data to test this hypothesis are lacking. In one Canadian sample of 66 individuals who self-reported childhood abuse, the Relationship Scales Questionnaire was administered along with a measure of post-traumatic stress symptoms (Reference Muller, Sicoli and LemieuxMuller et al, 2000). Only 24% of the sample were classified as having a secure attachment style. Those with preoccupied or fearful attachment reported the highest level of post-traumatic stress symptoms, and multiple regression analyses demonstrated the predictive power of a negative self-model, but not of a negative other-model. Theoretically, individuals with a negative internal working model of the self may use affect rather than cognition to guide their behaviour; faced with traumatic events, they may have difficulty in regulating ’raw’ emotions. The study by Muller et al did not test this point empirically. It was also a small study that was cross-sectional in design and was prone to recruitment bias, as volunteers had responded to flyers in the community. More recently, in a sample of 284 adults, attachment and dissociation were shown to mediate the relationship between childhood abuse and post-traumatic stress disorder following the terrorist attacks on New York’s World Trade Centre (Reference Twaite and RodriguezTwaite & Rodriguez, 2004).

Personality disorders

Attachment theory may provide significant insight into the developmental origins of personality disorders. Personality refers to enduring patterns of thought, motivation, emotional and impulse regulation, and interpersonal functioning. Attachment is related to these domains, albeit in the more specific context of personal relationships. Both personality disorders and attachment insecurity are associated with early adversity (Reference Meyer, Pilknos and ProlettiMeyer et al, 2001; Reference Nakash-Eisikovits, Dutra and WestenNakash-Eisikovits et al, 2002). With specific regard to borderline personality disorder (the personality disorder in which attachment has been most researched), Reference Dozier, Stovall, Albus, Cassidy and ShaverDozier et al(1999) note that:

Two large-scale studies have been conducted which show overlap between different personality disorders and attachment dimensions. In the first, Reference Brennan and ShaverBrennan & Shaver (1998) found in a non-clinical sample of 1407 undergraduates an association between attachment insecurity and self-reported personality disorders; of the 13 disorders studied, only psychopathy was unrelated to attachment insecurity. The study by Reference Fossati, Feeney and DonatiFossati et al(2003) assessed 487 psychiatric in-patients with a range of diagnoses. Through canonical correlation analysis, the data suggested that attachment avoidance might be associated with avoidant, depressive, paranoid and schizotypal personality disorders. Attachment anxiety might be associated with dependent, histrionic and borderline personality disorders. Similar but not identical results had been found in an earlier study of adolescent personality pathology (Reference Nakash-Eisikovits, Dutra and WestenNakash-Eisikovits et al, 2002).

Returning to Reference Fonagy, Leigh and SteeleFonagy et al(1996), the authors found that 75% of their patients diagnosed with DSM–III–R borderline personality disorder (27 out of 36) were classified as preoccupied on the Adult Attachment Interview, when the three-way classification (excluding U) was used (Table 2). Furthermore, a significant proportion received the sub-classification E3, which indicates ‘fearful preoccupation with traumatic events’. When the four-way classification (including category U) was used, nearly 90% (32 out of 36) of the sample with borderline personality disorder received an unresolved/disorganised U classification. The limitations of the study have been mentioned above.

In an Italian study of 40 patients with borderline personality disorder on a psychotherapy waiting list, Reference BaroneBarone (2003) found that 20 patients in her clinical sample (50%) were classified as unresolved/disorganised, compared with 3 out of 40 healthy controls (7%). Through sub-analysis of the Adult Attachment Interview sub-scales, she identified an ‘actively rejecting’ father and a ‘neglecting, poor-loving’ mother as especially relevant to the understanding of borderline personality disorder, and construed the personality disorder as a combination of maximising attachment strategies in the face of unresolved traumata.

Developmentally, unresolved loss or trauma in the parent predicts attachment disorganisation in the infant. Follow-up studies have shown that disorganised infants show ‘controllingness’ at 6–7 years of age, with both mothers and peers. Through role-reversal with their caregivers, these children may be providing a ’pseudo-secure base’ for themselves. At the same time, controlling children, often with helpless parents, are often unable to resolve frightening scenarios on picture completion tests. Reference HolmesHolmes (2004) hypothesises that:

This attachment-informed developmental model of borderline personality disorder needs to be tested with much more (longitudinal) research, and has important implications for psychotherapeutic treatment (Reference HolmesHolmes, 2004).

Eating disorders

Many explanations have been proposed for the aetiology of eating disorders, some of which have emphasised the importance of family factors. From an attachment point of view, eating disorders are theorised to be associated with deactivating strategies that minimise attachment needs. Eating disorders may be associated with a lack of attention to or inability to discriminate one’s inner distress cues, along with a focus on dieting and body appearance that may provide a diversion from such cues (Reference Cole-Detke and KobakCole-Detke & Kobak, 1996; Reference Dozier, Stovall, Albus, Cassidy and ShaverDozier et al, 1999). It has also been postulated that the symptoms of eating disorders maintain proximity to the attachment figure, albeit in a maladaptive way (Reference Orzolek-KronnerOrzolek-Kronner, 2002).

Empirical results to date have not consistently supported theory, which may again be a function of the differences across studies in the attachment measures/scoring systems, diagnostic criteria and samples used. The cross-sectional design of the key research on the topic needs to be borne in mind. As noted above, the study by Reference Cole-Detke and KobakCole-Detke & Kobak (1996) used the Adult Attachment Interview, but with the Q-set method of scoring, rather than the Main & Goldwyn system. The authors showed eating disorder symptoms (as opposed to the clinical syndromes) to be associated with attachment deactivation (dismissal) in a sample of college women, when depressive symptoms had been controlled for. On the other hand, Reference Fonagy, Leigh and SteeleFonagy et al(1996) found that 9 out of the 14 people (64%) with an eating disorder in their sample were preoccupied in their attachment in the three-way Main & Goldwyn classification; 13 out of the 14 (93%) showed attachment disorganisation in the four-way classification. No distinction was made between the different eating disorders.

In a clinical sample of 19 women (aged 15–46 years) with anorexia nervosa assessed using DSM–IV criteria and the Adult Attachment Interview, 15 were dismissing in their attachment (Reference Ward, Ramsay and TurnbullWard et al, 2001). Interestingly, although there was no association between the patients’ attachment patterns and those of their mothers, there was a high incidence of unresolved status in the latter group. The authors postulate that difficulty in emotional processing in these parents may have been transmitted to their daughters, serving as a risk factor for the development of anorexia nervosa.

Some research has examined the correlation between eating disorders and self-reported attachment. Using Bartholomew’s classification, Reference Broberg, Hjalmers and NevonenBroberg et al(2001) in Sweden compared a large sample of female out-patients who had eating disorders with a control group. Those with eating disorders were more insecure in their attachment. The authors further noted that the severity mattered more than the type of eating disorder in predicting the association between eating disorder symptomatology and self-rated attachment.