Modeling the hierarchical structure of attachment representations: A test of domain differentiation

Abstract

This research examined the hierarchical structure of attachment representations by directly measuring both global attachment orientations within distinct relationship domains (romantic, familial, and friendship), and attachment within a comprehensive range of specific relationships within each of these domains (e.g., current romantic partner, mother, best friend). In two independent samples, Hierarchical Linear Modeling analyses demonstrated that domain-specific representations were strongly associated with attachment ratings of specific relationships within that same domain (domain-relationship congruent associations) but were not (in general) associated with ratings of relationship-specific attachment in other domains (domain-relationship incongruent associations). These results provide evidence for the domain differentiation of multiple attachment representations – a defining feature of a hierarchically organized attachment representational network. Directions for future research integrating representations of attachment-related domains with the higher-order personality-esque component of the attachment representational hierarchy are briefly discussed.

Introduction

Individual differences in attachment have been defined and operationalized in numerous ways; as working models (Collins, Guichard, Ford, & Feeney, 2004), relational schema (Baldwin, 1992), appraisal and response components of a control systems model (Fraley & Shaver, 2000), and trait-like differences in the tendency to employ hyperactivating or deactivating behavioral strategies to regulate felt insecurity (Mikulincer & Shaver, 2003). Despite the subtleties of these theoretical differences, there exists a general consensus that (a) people hold multiple attachments toward a multitude of specific relationship partners, and (b) representations of these attachments are cognitively organized within a hierarchical structure (see Mikulincer and Shaver, 2003, Collins and Read, 1994). However, there has been surprisingly little empirical research directly modeling the structure of attachment representations (cf. Overall, Fletcher, & Friesen, 2003). The present research uses Hierarchical Linear Modeling (HLM) to test hypotheses regarding the differential associations between attachment measures that vary in specificity (or bandwidth) within a hierarchically organized attachment representational network.

HLM is uniquely suited for testing predictions regarding the magnitude of associations within hierarchical data structures. In the current context, HLM can first be used to partition the proportion of variance in participants’ attachment anxiety and avoidance within a range of specific relationships (e.g., ratings of attachment toward romantic partner, closest platonic friend, mother, father, and so forth) that is attributable to between- versus within-person variation. We refer to ratings of this type as relationship-specific, as they assess attachments within relationships with specific others. Second, HLM can then be used to examine the degree to which between-person variation in relationship-specific attachments is predicted by more global attachment representations of particular relationship domains. We use the term domain-specific to describe these representations, as such ratings refer to particular relationship domains or types (such as feelings typically experienced in romantic relationships in general), rather than to attachments toward specific persons. Thus, research using this approach can test the degree to which relationship-specific representations are coherently organized and explained by individual differences in more global attachment orientations versus the degree to which relationship-specific representations are the product of experiences and beliefs unique to that particular relationship.

La Guardia, Ryan, Couchman, and Deci (2000), for example, reported that approximately 25–35% of the variance in measures of anxiety and avoidance experienced within relationships with particular others (such as mother, father, romantic partner, roommate) reflected between-participant variation – that is, systematic variation indicative of individual differences that shape experiences across the multiple relationships experienced by each person. The remaining variance represented within-participant variation reflective of idiosyncratic differences within specific relationships unexplained by differences across individuals, including unidentified measurement error. Elaborating upon these findings, Pierce and Lydon (2001) reported that a global and non-specific measure of attachment orientation (specifically, how participants ‘felt in their close relationships, including their relationships to parents, siblings, close friends, relatives and their romantic partner’) predicted between-participant variation indicative of individual differences in attachment across specific relationships including mother, father, closest friend, and romantic partner. More recently, Sibley and Overall (2007) reported that domain-specific romantic attachment (assessing both previous and current romantic relationship experiences) was more strongly predictive of relationship-specific attachment ratings of particular romantic partners than it was of attachment orientations toward a range of non-differentiated close others with whom participants regularly interacted, including family members, friends and acquaintances.

These and related findings (e.g., Cozzarelli, Hoekstra, & Bylsma, 2000) demonstrate that people hold multiple attachment representations that differ in specificity, including representations specific to particular relationships (relationship-specific), as well as those that are more broad in bandwidth and reflect regularities in attachment orientation across relationships, including global representations that describe attachment across a range of relational contexts (e.g., Pierce & Lydon, 2001) and those that describe attachment in particular relationship domains (e.g., Sibley & Overall, 2007). These distinctions support, in part, a more complex theoretical model of the cognitive organization of the various attachment representations adults should possess. According to Collins and Read (1994), individuals will possess attachment representations pertaining to each of their specific relationships, which will be nested within domain-specific attachment representations that summarize regularities within distinct relational domains, including romantic, familial and friendship domains. These domain-specific representations, in turn, should be nested under a broad global representation reflecting attachment orientation across relationship contexts. The aforementioned research, however, has tested only specific components of this model and associations between all components of the model have not yet been tested within a single integrated study. Pierce and Lydon (2001), for instance, measured attachment toward close relationships in general rather than differentiating between relationship domains, whereas Sibley and Overall (2007) focused on romantic attachment and failed to measure familial and friendship representations.

Taking a somewhat different approach and assessing attachment across all three domains, Overall et al. (2003) reported that the covariance between relationship-specific attachment ratings toward specific romantic partners, specific family members, and specific close friends was best described by a three-level hierarchical structure in which attachments within specific relationships were nested under separate latent variables reflecting representations of each respective domain, which were in turn nested under a higher-order latent variable summarizing more global interpersonal orientations. In a separate series of analyses, domain-specific representations of romantic, familial, and friendship-based relationships were also found to all load on a higher-order latent variable indicative of a more global and non-specific attachment orientation. These findings support the important distinction between relationship-specific representations and representations that summarize more global but domain-based attachment orientations. Overall et al. (2003) did not, however, directly examine the associations between domain-specific measures and ratings of multiple relationships within each of the three domains.

How should the associations between domain-specific and relationship-specific attachment representations be structured? A hierarchically organized attachment representational network (as described above) provides the flexibility to operate in functional ways across relationship contexts, such that the activation of different representations should depend upon the specificity of the relationship context (Collins and Read, 1994, Overall et al., 2003). Elaborating on this premise, Sibley and Liu (2006) proposed a context-congruence hypothesis of attachment effects, predicting that representations of particular relationships and relationship domains will exert their strongest effects on interpersonal behaviors, cognitions, and emotions when the specific relational context corresponds to the domain that representation refers to. Thus, attachment orientations within similar types of relationship should be more closely associated because they are more likely to encompass similar interpersonal experiences and govern attachment system functioning in similar contexts (Mikulincer & Shaver, 2003). Taking the romantic relationship context, for example, relationship-specific representations of particular romantic partners (both past and current) should all be closely associated. In addition, attachment representations of current and past romantic partners should be closely linked with a more general domain-specific orientation within the romantic domain but be less closely associated with attachment representations of the familial and friendship domains.

Put simply, a hierarchically organized attachment network would produce a specific pattern of differential associations across measures of domain-specific and relationship-specific attachment. Specifically, domain-specific representations should be strongly associated with attachment ratings of specific relationships within that same domain (domain-relationship congruent associations) but should not be associated with ratings of relationship-specific attachment in other domains (domain-relationship incongruent associations). One alternative possibility, for instance, is that attachments within specific relationships might be more (or as) strongly predicted by domain-specific romantic attachment because current romantic partners are the most central attachment figure for the majority of adults (Trinke & Bartholomew, 1997) and, thus, representations within the romantic domain might be the most chronically accessible. Alternatively, perhaps attachment is strongly consistent across domains so that relationship-specific ratings are predicted equally or at least significantly by all three global attachment orientations. This pattern of associations would suggest that the hypothesized domain-differentiated mid-level tier of the representational network should simply give way to the more global representation of attachment across relational contexts. Although we think the domain differentiated pattern described above is most likely based on prior empirical and theoretical work, previous research has not directly examined the extent to which attachment ratings of the romantic, familial, and friendship domains differentially predict multiple ratings of specific relationships within each of these domains.