Methodological Insights

Several of the articles contained in this Special Issue provide direct evidence that the study of attachment in the context of atypical caregiving both presents significant methodological challenges and ultimately generates valuable opportunities to acquire insight into the nature of individual differences, especially for lower base rate attachment-related phenomena (e.g., unresolved and preoccupied states of mind). For example, Poehlmann-Tynan et al.’s (2017) study of child attachment in the context of parental incarceration shows clearly that the development of new observational procedures and coding systems is often a prerequisite to studying attachment in high-risk contexts given the extra-scientific constraints that bedevil scholars in this area. Likewise, two other papers contained in this Special Issue describe efforts to design or further validate assessments of the quality of adolescent–parent attachments via direct observations (Kobak et al., Reference Kobak, Zajac, Abbott, Zisk and Bounoua2017; Martin, Sturge-Apple, et al., Reference Martin, Sturge-Apple, Davies, Romero and Buckholz2017). Such methodological contributions to the literature are timely and welcome.

Perhaps even more notable in terms of their methodological significance for attachment research, however, are the studies reported in this Special Issue that address unresolved yet fundamental questions about how individuals vary with respect to how they describe their early experiences with primary caregivers in the context of the Adult Attachment Interview (AAI; George, Kaplan, & Main, Reference George, Kaplan and Main1984–1996). The canonical view of individual differences in adults’ narratives about their childhood experiences is built on two distinct sets of assumptions embedded in the Main and Goldwyn (Reference Main and Goldwyn1984–1998) coding system for the AAI (see Main, Goldwyn, & Hesse, Reference Main and Goldwyn2003–2008). The first of these assumptions is that individual differences in adult attachment are best conceptualized as categorically rather than continuously distributed in the population. Arguably even more important, however, is a second set of assumptions regarding the primary axes on which adults vary while talking about their childhood attachment experiences. More specifically, Main and Goldwyn's coding system for the AAI implies that security in adulthood simultaneously reflects the capacity to (a) tell an internally consistent narrative about one's childhood experiences while (b) not becoming emotionally overwrought when talking about those experiences, as reflected either in active or passive preoccupation. In this traditional account that is now well enshrined in the developmental literature, insecurity, by contrast, is conceptualized as taking one of two mutually incompatible organized forms (i.e., dismissing or preoccupied states of mind, but not both).

Prior research based on large, normative-risk cohorts appear inconsistent with both of these sets of assumptions. First, in the only two existing taxometric studies of the AAI that predate this Special Issue (i.e., Fraley & Roisman, Reference Fraley and Roisman2014; Roisman, Fraley, & Belsky, Reference Roisman, Fraley and Belsky2007), dismissing states of mind during the AAI (reflected in the normalization of objectively harsh childhood experiences and/or the idealization of childhood experiences with primary caregivers) were demonstrated to be continuously rather than categorically distributed (evidence regarding the taxonicity of preoccupation was indeterminate). Second, all existing factor analytic evidence from a variety of attachment laboratories across the world suggests that, rather than being mutually incompatible, dismissing and preoccupied states of mind appear to be essentially orthogonal (for reviews, see Roisman & Booth-LaForce, Reference Roisman and Booth-LaForce2014; Martin, Bureau, et al., Reference Martin, Bureau, Lafontaine, Cloutier, Hsiao, Pallanca and Meinz2017). Perhaps even more surprising to many attachment scholars, these same factor analyses have demonstrated consistently that indicators of unresolved states of mind (reflecting psychological confusion during the AAI about loss and/or abuse-related experiences specifically) and indicators of preoccupation about early caregiving experiences generally (i.e., passive and/or angry discourse during the AAI) load on a single factor, suggesting that these are not psychologically distinct phenomena, as has been so widely assumed.

Van IJzendoorn and Bakermans-Kranenburg (Reference van IJzendoorn and Bakermans-Kranenburg2014) observed several years ago that a potentially significant limitation of prior taxometric and factor analytic work focused on individual differences in AAI narratives is the lower risk nature of the samples that have been featured in most such psychometric reports (see Roisman, Fraley, & Booth-LaForce, Reference Roisman, Fraley and Booth-LaForce2014). At the core of van IJzendoorn and Bakermans-Kranenburg's (Reference van IJzendoorn and Bakermans-Kranenburg2014) critique is the descriptive insight that preoccupied and unresolved states of mind represent relatively low base rate phenomena, especially in typical-risk cohorts, and that any differences between these groups (assuming they are true taxa) might well be obscured in samples that contain relatively few prototypically preoccupied and/or unresolved participants.

It is in this context that several of the studies reported in this Special Issue take on notable significance. Even if we set aside the additional taxometric and confirmatory factor analytic evidence pertinent to the AAI reported by Raby, Yarger, et al. (Reference Raby, Yarger, Lind, Fraley, Leerkes and Dozier2017; given that many of the participants in that impressively large sample study were at relatively low risk), Martin, Bureau, et al. (Reference Martin, Bureau, Lafontaine, Cloutier, Hsiao, Pallanca and Meinz2017) and Raby, Labella, et al. (Reference Raby, Labella, Martin, Carlson and Roisman2017) both present factor analyses demonstrating in higher risk cohorts once again that (a) AAI state of mind scales were most parsimoniously represented by two weakly correlated dismissing and preoccupied state of mind dimensions and (b) unresolved and preoccupied AAI states of mind loaded on a common factor in both samples. Indeed, the Raby, Labella, et al. (Reference Raby, Labella, Martin, Carlson and Roisman2017) report, drawing on data from the landmark Minnesota Longitudinal Study of Risk and Adaptation (Sroufe, Egeland, Carlson, & Collins, Reference Sroufe, Egeland, Carlson and Collins2005), actually extended this two-factor structure to the Current Relationship Interview (Crowell & Owens, Reference Crowell and Owens1996), a close analogue of the AAI focused on experiences with a current romantic partner.

Does the addition to the literature of the factor analyses of the AAI and Current Relationship Interview reported in this Special Issue effectively put an end to debate about the latent structure of individual differences in the ways in which adults discuss their childhood and current attachment experiences? Of course not, but such work definitely does shift the burden of evidence to those who persist in using coding systems in ways that assume that (a) attachment is categorically distributed in adulthood, (b) dismissing and preoccupied states of mind are mutually incompatible, and/or (c) preoccupied and unresolved states of mind represent psychologically distinct attachment phenomena. The empirical support for each of these long-held assertions, at least in regard to evidence derived from taxometric and/or factor analyses of AAI state of mind data drawn from both normative and higher risk cohorts, is largely nonexistent in the developmental literature.

Biological Mechanisms

Another exciting aspect of the work presented in this Special Issue is the relatively large number of reports emphasizing the potential value of biological markers in informing both (a) the origins of attachment-related variation (e.g., Mulder et al., Reference Mulder, Rijlaarsdam, Luijk, Verhulst, Felix, Tiemeier and van IJzendoorn2017) and (b) the legacy of early attachment-relevant experiences (Blaze & Roth, Reference Blaze and Roth2017; Fearon et al., Reference Fearon, Tomlinson, Kumsta, Skeen, Murray, Cooper and Morgan2017; Quevedo et al., Reference Quevedo, Waters, Scott, Roisman, Shaw and Forbes2017; Thijssen et al., Reference Thijssen, Muetzel, Bakermans-Kranenburg, Jaddoe, Tiemeier, Verhulst and van IJzendoorn2017). As just a few salient examples, Blaze and Roth (Reference Blaze and Roth2017) use a rodent model to extend their path-finding work demonstrating the role of early maltreatment in abnormal patterns of methylation in the brain-derived neurotrophic factor (Bdnf) gene in the medial prefrontal cortex; and, in the human context, both Thijssen et al. (Reference Thijssen, Muetzel, Bakermans-Kranenburg, Jaddoe, Tiemeier, Verhulst and van IJzendoorn2017) and Quevedo et al. (Reference Quevedo, Waters, Scott, Roisman, Shaw and Forbes2017) make use of functional magnetic resonance imaging to study the ways in which early caregiving experiences are reflected in patterns of neural activation. Such work has the potential to advance behavioral science in part by identifying some of the specific mechanisms that might account for now well-established associations observed between early attachment experiences with primary caregivers and aspects of (mal)adjustment over the early life course (Groh, Fearon, van IJzendoorn, Bakermans-Kranenburg, & Roisman, Reference Groh, Fearon, van IJzendoorn, Bakermans-Kranenburg and Roismanin press).

These innovative and diverse papers on biological correlates of attachment-related individual differences surely suggest many more questions than they can possibly answer definitively. For example, only critical replication and extension efforts will ultimately be dispositive regarding which subset of the many, diverse biological systems and methods leveraged in the reports featured here will prove most reliably informative about the legacy of early attachment experiences. Regardless, we anticipate that these studies will serve the field well by providing additional direction for attachment scholarship seeking to understand biological level of analysis mechanisms, an important focus that nonetheless should not be emphasized to the exclusion of a broader search for other relevant developmental mechanisms (e.g., those at the cognitive and behavioral levels of analysis).

Attachment Interventions

Of the 24 articles in this Special Issue, it is noteworthy that 10 feature the results of attachment-relevant interventions. That so many investigators are now actively engaged in work designed to translate the basic science of attachment in the interest of serving the higher risk populations in greatest need for such programs is laudable and very much in keeping with a theory that got its start in clinical observations (Bowlby, Reference Bowlby1951). Such work is likewise important in light of its potential to improve our basic science via enhanced causal inferences, at least when implemented using robust experimental designs (i.e., randomized control trials).

The research featured in this Special Issue clearly pushes the boundaries of the status quo of attachment-based intervention in a number of potentially generative ways. Three of the reports find efficacy of attachment-related intervention not merely for interpersonal outcomes (as might be expected) but for improving aspects of cognitive development (Cassidy et al., Reference Cassidy, Brett, Gross, Stern, Martin, Mohr and Woodhouse2017; Dubois-Comtois et al., Reference Dubois-Comtois, Cyr, Tarabulsy, St.-Laurent, Bernier and Moss2017; Lind, Raby, Caron, Roben, & Dozier, Reference Lind, Raby, Caron, Roben and Dozier2017). Another interesting report based on a French sample hones in on some of the specific interpersonal behavioral change mechanisms (i.e., improvements in disrupted maternal communication) by which attachment-informed home visitations offset risk for disorganized attachment (Tereno et al., Reference Tereno, Madigan, Lyons-Ruth, Plamondon, Atkinson, Guedeney and Guedeney2017). Yet another paper reports that children abandoned in Romania but subsequently randomized to high-quality foster care (compared to children who experienced care as usual) showed fewer signs of reactive attachment disorder and disinhibited social engagement years later (around age 12 years; Humphreys et al., Reference Humphreys, Nelson, Fox and Zeanah2017).

That said, the intervention studies presented in this Special Issue also suggest that there is a good deal of progress yet to be made in this area. One broad issue that would benefit from additional attention in future intervention work of this kind is that of study design. First, even the largest interventions reported in this Special Issue (e.g., Cassidy et al., Reference Cassidy, Brett, Gross, Stern, Martin, Mohr and Woodhouse2017; Humphreys et al., Reference Humphreys, Nelson, Fox and Zeanah2017; Lind et al., Reference Lind, Raby, Caron, Roben and Dozier2017) consisted of fewer than 200 participants, with adequate power to detect only medium or larger population effects. In contrast, many of the studies reported here included approximately 100 participants, and a handful of “proof of concept” reports were based on fewer than 50 individuals (e.g., Dubois-Comtois et al., Reference Dubois-Comtois, Cyr, Tarabulsy, St.-Laurent, Bernier and Moss2017; Schact et al., Reference Schacht, Meins, Fernyhough, Centifanti, Bureau and Pawlby2017; Swain et al., Reference Swain, Ho, Rosenblum, Morelen, Dayton and Muzik2017). Intervention research presents inherent challenges and is immensely expensive and time consuming, but the point remains that only large sample trials are capable of detecting the kinds of small (though ideally enduring) differences between intervention and control groups that might be reasonably expected in this area. Equally important, small sample interventions, combined with a large numbers of potential outcome variables that might be examined, create excessive risk for Type 1 errors. In short, if we are to be well positioned to produce replicable answers to important questions about to what extent attachment interventions work, sample sizes in future intervention studies will need to reflect the statistical power requirements necessary to produce relatively precise estimates of focal effects, or else we risk false discovery.

Second, those who read this Special Issue will note a great deal of variability in the kinds of specific attachment-relevant intervention strategies deployed across reports, ranging from interpersonal psychotherapy administered to depressed mothers (Handley et al., Reference Handley, Michl-Petzing, Rogosch, Cicchetti and Toth2017) to home visitation (Tereno et al., Reference Tereno, Madigan, Lyons-Ruth, Plamondon, Atkinson, Guedeney and Guedeney2017) to attachment-based video feedback (Dubois-Comtois et al., Reference Dubois-Comtois, Cyr, Tarabulsy, St.-Laurent, Bernier and Moss2017) to a wide range of manualized yet distinct attachment intervention programs, including the Attachment and Biobehavioral Catch-up for Toddlers Intervention (Lind et al., Reference Lind, Raby, Caron, Roben and Dozier2017), the Circle of Security Parenting Intervention (Cassidy et al., Reference Cassidy, Brett, Gross, Stern, Martin, Mohr and Woodhouse2017), and Mothering From the Inside Out (Suchman et al., Reference Suchman, DeCoste, McMahon, Dalton, Mayes and Borelli2017). Indeed, there were as many distinct approaches to attachment-related intervention reported in this Special Issue as there were laboratories that contributed intervention data.

Of course, much of the variation just mentioned reflects intentional efforts to optimally meet the distinct needs of the at-risk groups that are the focus of the various intervention efforts described in this issue (e.g., Humphreys et al., Reference Humphreys, Nelson, Fox and Zeanah2017). Nonetheless, developing a cumulative science in this area will require that we continue to identify which of these approaches to attachment-related intervention are uniquely efficacious, and for whom (Bakermans-Kranenburg, van IJzendoorn, & Juffer, Reference Bakermans-Kranenburg, van IJzendoorn and Juffer2003). These are nontrivial questions, particularly given the null effects reported in at least one of the larger trials published in the Special Issue (Cassidy et al., Reference Cassidy, Brett, Gross, Stern, Martin, Mohr and Woodhouse2017). One potentially productive way forward is for researchers in this area to not only include active (nonattachment focused) control groups in their intervention studies (see, e.g., Lind et al., Reference Lind, Raby, Caron, Roben and Dozier2017; and Suchman et al., Reference Suchman, DeCoste, McMahon, Dalton, Mayes and Borelli2017, for excellent examples of this practice) but also to commit to collaborative, large sample trials that directly compare the efficacy of some of the most well-developed attachment interventions now available. Such work will require a great deal of cooperation among investigators, but also has great potential to most efficiently discover how best to improve the lives of children and adults via attachment-related intervention.