New directions in the examination of IWM formation during infancy

Despite Bowlby's hypothesis that infants develop IWMs during the first year of life (see also Main et al., Reference Main, Kaplan and Cassidy1985), almost no empirical work has focused on attachment representations during infancy (instead, most research on IWMs has involved children, adolescents, and adults). We believe, as do others (Johnson et al., Reference Johnson, Dweck, Chen, Stern, Ok and Barth2010; Sherman & Cassidy, Reference Sherman and Cassidy2013; Thompson, Reference Thompson, Cassidy and Shaver2008), that IWMs can be studied in infancy. Such work is made possible by recent efforts to bridge social–emotional and cognitive developmental research (e.g., Calkins & Bell, Reference Calkins and Bell2010; Olson & Dweck, Reference Olson and Dweck2008), along with methodological advances and accumulating research on an array of previously unexplored infant mental capacities.

Attachment researchers have assumed that infants recall the emotional nature of their attachment-related social experiences with specific individuals (e.g., experiences of comfort with vs. rejection by mother), and that they use these memories to create IWMs that guide their attachment behavior in subsequent interactions with these individuals. This claim has been supported with correlational research findings; for example, observations indicate that infants' daily interactions with attachment figures are linked to their IWMs reflected in behavior in the Strange Situation (Ainsworth et al., Reference Ainsworth, Blehar, Waters and Wall1978). These findings can now be supplemented with results from experimental studies.

There is a compelling body of experimental work showing that infants extract complex social–emotional information from the social interactions they observe. For example, they notice helpful and hindering behaviors of one “person” (usually represented by a puppet or a geometric figure) toward another, they personally prefer individuals who have helped others, they form expectations about how two characters should behave toward each other in subsequent interactions, and they behave positively or negatively toward individuals based on what they have observed (e.g., Hamlin & Wynn, Reference Hamlin and Wynn2011; Hamlin, Wynn, Bloom, & Mahajan, Reference Hamlin, Wynn, Bloom and Mahajan2011). This work could and should be extended to include attachment relationships, revealing in detail how infants form “models” of particular adults and then modify their emotional reactions and social behaviors toward those adults accordingly (Johnson et al., Reference Johnson, Dweck, Chen, Stern, Ok and Barth2010). At present, there is no experimental research showing that infants form expectations about the later social behavior of another person toward them based on the infants' own past interactions with that person, a capacity that is assumed to underlie infants' development of working models of their caregivers.

As explained in detail in another paper (Sherman & Cassidy, Reference Sherman and Cassidy2013), we urge infancy researchers to consider the specific cognitive and emotional capacities required to form IWMs and then to examine these capacities experimentally. Methods used by researchers who study infant cognition, but rarely used by attachment researchers (e.g., eye-tracking, habituation paradigms), will prove useful. For example, habituation paradigms could allow attachment researchers to study infant IWMs of likely mother and infant responses to infant distress (see Johnson et al., Reference Johnson, Dweck, Chen, Stern, Ok and Barth2010).

Another research area relevant to attachment researchers' conception of IWMs concerns infants' understanding of statistical probabilities. When considering individual differences in how mothering contributes to attachment quality, Bowlby (Reference Bowlby1969/1982) adopted Winnicott's (Reference Winnicott1953) conception of “good enough” mothering; that is, mothering that assures a child that probabilistically, and often enough, the mother will prove responsive to the child's signals. Implicit in such a perspective is the assumption that an infant can make probabilistic inferences. Only recently has there been a surge in interest in the methods available to evaluate this assumption of attachment theory (e.g., Krogh, Vlach, & Johnson, Reference Krogh, Vlach and Johnson2013; Pelucchi, Hay, & Saffran, Reference Pelucchi, Hay and Saffran2009; Romberg & Saffran, Reference Romberg and Saffran2013; Xu & Kushnir, Reference Xu and Kushnir2013).

One useful conceptual perspective, called rational constructivism, is based on the idea that infants use probabilistic reasoning when integrating existing knowledge with new data to test hypotheses about the world. Xu and Kushnir (Reference Xu and Kushnir2013) reviewed evidence that, by 18 months of age, infants use probabilistic reasoning to evaluate alternative hypotheses (Gerken, Reference Gerken2006; Gweon, Tenenbaum, & Schulz, Reference Gweon, Tenenbaum and Schulz2010), revise hypotheses in light of new data (Gerken, Reference Gerken2010), make predictions (Denison & Xu, Reference Denison and Xu2010), and guide their actions (Denison & Xu, Reference Denison and Xu2010). Moreover, infants are capable of integrating prior knowledge and multiple contextual factors into their statistical computations (Denison & Xu, Reference Denison and Xu2010; Teglas, Girotto, Gonzales, & Bonatti, Reference Teglas, Girotto, Gonzales and Bonatti2007; Xu & Denison, Reference Xu and Denison2009). Xu and Kushnir (Reference Xu and Kushnir2013) have further proposed that these capacities appear to be domain-general, being evident in a variety of areas: language, physical reasoning, psychological reasoning, object understanding, and understanding of individual preferences. Notably absent from this list is the domain of social relationships, including attachment relationships.

Several questions about probabilistic inferences can be raised. Do infants make such inferences about the likely behavior of particular attachment figures, and could this ability account for qualitatively different attachments to different individuals (e.g., mother as distinct from father)? Do infants use probabilistic reasoning when drawing inferences related to the outcomes of their own attachment behaviors? (This is related to if–then contingencies: “If I cry, what is the probability that x will occur?”) How complex can this infant reasoning become and across what developmental trajectory? “If I do x, the likelihood of outcome y is 80%, but if I do w, the likelihood of y is only 30%.” Do infants consider context? “If I do x, the likelihood of y is 90% in context q, but only 20% in context r.” How do infants calculate variability in these probabilities across attachment figures?

In sum, it seems likely that infants use statistical inference to understand their social worlds. This ability would seem to be evolutionarily adaptive in relation to attachment figures because infants could incorporate probabilistic inferences into their IWMs and use them to guide their attachment behavior. Important advances in our understanding of attachment behavior might occur with respect to how and when this incorporation happens and also with respect to the role of statistical inference in infants' openness to change in response to changing environmental input (e.g., in response to interventions designed to change parental behavior).

Another level of “representation” or internal structure: Physiology

Since the time of Bowlby's original writings, one important advance that has extended our understanding of the link between attachment and response to threat has roots in Myron Hofer's laboratory in the 1970s. Hofer, a developmental psychobiologist, noticed defensive vocal protest responses to maternal separation in infant rat pups and asked what nonrepresentational process could account for them. He and his colleagues conducted a series of tightly controlled experiments to identify what physiological subsystems, which he called hidden regulators, are disrupted when mothers are removed from their pups (for reviews, see Hofer, Reference Hofer2006; Polan & Hofer, Reference Polan, Hofer, Cassidy and Shaver2008). The pups exhibit changes in multiple physiological and behavioral systems, such as those controlling heart rate, body temperature, food intake, and exploration. Hofer concluded that mother–infant interactions have embedded within them a number of vital physiological regulatory functions that are disrupted by separation from mother and do not require cognitive mediators. These regulators can be disentangled by experimentally manipulating parts of a “mother”: the food she provides, her warmth, her licking and grooming, and so forth. Later, Meaney and colleagues (e.g., Liu et al., Reference Liu, Diorio, Tannenbaum, Caldji, Francis and Freedman1997; reviewed in Meaney, Reference Meaney2001) found that rat pups that received high levels of maternal licking and grooming and arched-back nursing positions had milder responses to threat and increased exploratory behavior, effects that lasted into adulthood. (The effects lasted into subsequent generations as a function of maternal affection in each successive generation.) This research group further found that individual differences in maternal behavior were mediated by differences in offsprings' gene expression (Weaver et al., Reference Weaver, Cervoni, Champagne, D'Alessio, Sharma and Seckl2004), a finding that has opened up a new research domain for researchers studying both animals and humans (Sharp et al., Reference Sharp, Pickles, Meaney, Marshall, Tibu and Hill2012; Suomi, Reference Suomi2011).

Early attachment-related experiences and human infant biological response to stress

In humans, a fully developed stress response system, the hypothalamic–pituitary–adrenal (HPA) axis, is present at birth (Adam, Klimes-Dougan, & Gunnar, Reference Adam, Klimes-Dougan, Gunnar, Coch, Dawson and Fischer2007). A growing body of research indicates that differences in the quality of early care contribute to variations in the initial calibration and continued regulation of this system. This regulation in turn plays an important role in shaping behavioral responses to threat (Jessop & Turner-Cobb, Reference Jessop and Turner-Cobb2008).

Researchers have examined connections between caregiving experiences and infant stress physiology by comparing infants' cortisol levels before and after a stressful task (e.g., the Strange Situation). For example, Nachmias, Gunnar, Mangelsdorf, Parritz, and Buss (Reference Nachmias, Gunnar, Mangelsdorf, Parritz and Buss1996) found that inhibited toddlers who were insecurely attached to their caregivers exhibited elevated cortisol levels following exposure to novel stimuli. There is also experimental evidence that mothers' touch buffers infants' cortisol stress response (in this case, during the Still Face laboratory procedure, in which mothers are asked to cease interacting emotionally with their infants; Feldman, Singer, & Zagoory, Reference Feldman, Singer and Zagoory2010). Children living in violent families endure particularly stressful caregiving environments, which are extremely dysregulating for them (Taylor, Repetti, & Seeman, Reference Taylor, Repetti and Seeman1997). A number of studies have documented the disrupted stress response of maltreated children (e.g., De Bellis et al., Reference De Bellis, Baum, Birmaher, Keshavan, Eccard and Boring1999; Hart, Gunnar, & Cicchetti, Reference Hart, Gunnar and Cicchetti1995). Even living in a family in which the violence does not involve them directly has negative consequences for children, and studies suggest that the quality of caregiving in these harsh environments plays an important role in modifying the stress response (e.g., Hibel, Granger, Blair, Cox, & the FLP Investigators, Reference Hibel, Granger, Blair and Cox2011).

Attachment as a regulator of infant stress reactivity: Further questions

Just as infants are thought to have evolved a capacity to use experience-based information about the availability of a protective caregiver to calibrate their attachment behavioral system (Main, Reference Main1990), and given the close intertwining of the attachment and fear systems, it is likely that infants also evolved a capacity to use information about the availability of an attachment figure to calibrate their threat response system at both the behavioral and physiological levels (Cassidy, Reference Cassidy2009). This capacity is probably not solely cognitive, which raises important questions for research: How are representational and physiological processes linked and how do they influence each other and affect child functioning? Does the nature of their interaction vary across particular aspects of child functioning and across developmental periods? How can we understand these interactions in relation to both normative development and individual differences?

In humans, representations and physiological (e.g., stress) reactions are thought to affect each other in ways unlikely to occur in other species. Sapolsky (Reference Sapolsky2004) noted that, in humans, representational processes (the anticipation of threat when none currently exists) can launch a stress response. Relatedly, Bowlby (Reference Bowlby1973), focusing on the link between attachment and fear, specified representational “forecasts of availability or unavailability” of an attachment figure as “a major variable that determines whether a person is or is not alarmed by any potentially alarming situation” (p. 204). Thus, the representations that others will be unavailable or rejecting when needed (i.e., representations that characterize insecure attachment) could contribute to chronic activation of physiological stress response systems, as could the associated representations of others as having hostile intentions (Dykas & Cassidy, Reference Dykas and Cassidy2011). Conversely, in times of both anticipated and actual threat, the capacity to represent a responsive attachment figure can diminish physiological responses associated with threatening or painful experiences (see Eisenberger et al., Reference Eisenberger, Master, Inagaki, Taylor, Shirinyan and Lieberman2011; Coan, Schaefer, & Davidson, Reference Coan, Schaefer and Davidson2006). Moreover, consideration of linkages between representational and nonrepresentational processes must include the possibility that causality flows in both directions: physiological stress responses can presumably prompt a person to engage in higher level cognitive processes to understand, justify, or eliminate the stressor.

When and how do young children use attachment-related representations as regulators of stress? Neither normative trajectories nor individual differences in the use of representations to influence stress reactivity have been examined extensively. Evidence that stress dysregulation can lead to the conscious engagement of representational processes comes from children as young as 4 who are able to describe strategies for alleviating distress (e.g., changing thoughts, reappraising the situation, mental distraction; Sayfan & Lagattuta, Reference Sayfan and Lagattuta2009). Less studied but of great interest are possible “automatic emotion regulation” processes (Mauss, Bunge, & Gross, Reference Mauss, Bunge and Gross2007) that do not involve conscious or deliberate self-regulation. Recent studies of adults show that there are such processes, that there are individual differences in them that might relate to attachment orientations, that they are associated with particular brain regions that are not the same as those associated with conscious, deliberate emotion regulation, and that they can be influenced experimentally with priming procedures.

Many researchable questions remain: given the extent to which many forms of psychopathology reflect problems of self-regulation in the face of stress (e.g., Kring & Sloan, Reference Kring and Sloan2010), can “hidden regulators” stemming from infant–mother interactions tell us about the precursors of psychopathology? What about hidden regulators embedded within a relationship with a therapist (who, according to Bowlby, Reference Bowlby1988, serves as an attachment figure in the context of long-term psychotherapy)? When change occurs following long-term therapy, does this change emerge through cognitive representations, changes at the physiological level, or both? See Cassidy et al. (Reference Cassidy, Ehrlich, Sherman, Mikulincer and Shaver2013) for additional suggestions for future research.

A focus on secure base provision

For Bowlby (Reference Bowlby1988), the secure base concept was the heart of attachment theory: ‘‘No concept within the attachment framework is more central to developmental psychiatry than that of the secure base’' (pp. 163–164). When parents provide a secure base, their children's confidence in the parents' availability and sensitive responsiveness when needed allows the children to explore the environment freely. The secure base phenomenon contains two intertwined components: a secure base from which a child can explore and a haven of safety to which the child can return in times of distress. As noted earlier, the central cognitive components of secure attachment are thought to reside in a secure base script (i.e., a script according to which, following a distressing event, the child seeks and receives care from an available attachment figure, experiences comfort, and returns to exploration).

If the goal of research is to understand what components of a parent's behavior allow a child to use the parent as a secure base, researchers should focus as precisely as possible on the parent's secure base provision rather than on his or her parental behavior more broadly. Through experience-based understanding of parental intentions and behavior, an infant gathers information to answer a central secure base question: what is my attachment figure likely to do when activation of my attachment system leads me to seek contact? If experiences lead the infant to believe that the parent will be responsive (most of the time) to behaviors related to activation of his/her attachment system, then the infant will use the parent as a secure base, and behavioral manifestations of the secure base script will appear (i.e., the secure base script will be evident in the Strange Situation attachment assessment, and the infant will be classified as secure). In Reference Waters and Cummings2000, Waters and Cummings, when proposing an agenda for the field in the millennium of the 2000s, urged that the secure-base concept be kept “at center stage in attachment theory and research” (p. 164). We share this opinion and believe that additional consideration of the secure base notion will provide a useful framework within which to consider parental behavior as a predictor of infant attachment.

Bowlby (Reference Bowlby1988) emphasized that an infant's sense of having a secure base resides in the infant's confidence that parental sensitive responsiveness will be provided when needed (e.g., specifying “especially should [the infant] become tired or frightened,” p. 132). As such, it may be useful for attachment researchers to frame their question as the following: which contexts provide the infant with information about the parent's likely behavior when needed, not in all contexts, but specifically in response to activation of the infant's attachment system? Bowlby (Reference Bowlby1969/1982) described the relevant contexts as “fall[ing] into two classes: those that indicate the presence of potential danger or stress (internal or external) and those concerning the whereabouts and accessibility of the attachment figure” (p. 373).

Especially during the early years of life, both of these circumstances are likely to be associated with infant distress. This association has led some writers to wonder whether maternal response to infant distress is particularly predictive of infant attachment quality (e.g., Thompson, Reference Thompson1997), and there is compelling evidence that this is the case (e.g., Del Carmen, Pedersen, Huffman, & Bryan, Reference Del Carmen, Pedersen, Huffman and Bryan1993; Leerkes, Reference Leerkes2011; Leerkes, Parade, & Gudmundson, Reference Leerkes, Parade and Gudmundson2011; McElwain & Booth-LaForce, Reference McElwain and Booth-LaForce2006). When infants experience comfort from parental sensitive responses to their distress, they develop mental representations that contribute to security (“When I am distressed, I seek care, and I am comforted”). These representations are then thought to guide secure attachment behavior, and the physiological regulation that comes from regaining calmness in contact with the parent is thought to calibrate the child's stress reactivity systems and feed back into further secure mental representations (e.g., Cassidy et al., Reference Cassidy, Ehrlich, Sherman, Mikulincer and Shaver2013; Suomi, Reference Suomi, Cassidy and Shaver2008). The greater predictive power of the maternal response to distress, compared to maternal response to nondistress, may emerge from the considerable intertwining of infant distress and the infant's attachment system during the first year of life.

Future studies attempting to predict infant attachment might benefit from a framework that considers two components of parental behavior: (a) parental behavior related specifically to the secure base function of the infant's attachment system as Bowlby described it (see above), and (b) parental response to infant distress. Table 1 presents a 2 (attachment-related or not) × 2 (infant distressed or not) matrix that gives rise to a number of research questions. One key question is the following: is parental behavior in response to an infant's attachment behavioral system most predictive of infant attachment, regardless of whether the infant is distressed (i.e., parental behavior in both Cells 1 and 2)? Another set of questions relates to distress: is parental response to any form of infant distress the most central predictor of infant attachment (i.e., parental behavior in both Cells 1 and 3)? Does the termination of the physiological and emotional dysregulation of distress, no matter what the cause, that occurs through parental care solidify a tendency to use the parent as a secure base, or do the cognitive models that derive from experiences of distress in different contexts (e.g., distress during play vs. distress when seeking comfort) contribute differentially to secure base use? Most previous research has not drawn distinctions concerning the context of infant distress; future work that considers this distinction is needed.

Table 1.

Role of distress and context in parenting

Note: The following examples describe 5- to 12-month-old infants participating in studies with their mothers in Cassidy's lab. Cell 1: The context is attachment related, and the infant is distressed: after having been left alone in an unfamiliar laboratory playroom, a crying 12-month-old crossed the room to her returning mother and reached to be picked up. Cell 2: The context is attachment related, and the infant is not distressed: an 8-month-old infant had been playing contentedly for 20 min near her mother at home. The mother had been sitting on the floor holding a toddler whose hair she was braiding. When the mother finished and the toddler moved away, the infant crawled to the mother, clambered up on her lap, and snuggled in for a hug; after exchanging tender pats with her mother, the infant returned to play on the floor. The lack of accessibility to the mother may have led to the infant's seeking contact in a manner that did not involve other activities (e.g., play or feeding). Cell 3: The context is not attachment related, and the infant is distressed: a 12-month-old infant became distressed when a toy was removed. Cell 4: The context is not attachment related, and the infant is not distressed: an infant, with her mother nearby, played happily with toys.

^aIt is important to keep in mind that Bowlby (Reference Bowlby1969/1982) believed that the attachment system is best viewed as being “never idle” (p. 373), with a continuum of infant behavioral manifestations ranging from the simple monitoring of levels of threat and attachment figure availability to the high distress and intense attachment behavior evident when a crying infant rushes to be picked up. Attachment system related indicates that the context in question is related centrally to the infant's attachment behavioral system rather than to another behavioral system.

Additional questions raised by Table 1 include, is it the combination of maternal behavior when the infant's attachment system is central, along with any behavioral response to infant distress, that best predicts infant attachment (i.e., maternal behavior in Cells 1, 2, and 3)? Finally, is it the case (as some have suggested; e.g., Pederson & Moran, Reference Pederson and Moran1999) that maternal behavior in all four cells is predictive of infant attachment? Attempts to increase understanding of the precursors of infant attachment will require the development of detailed coding systems.

Finally, it will be crucial for future research conducted within a secure base framework to identify the specific maternal behaviors in response to activation of the infant's attachment system that predict infant security (for one approach, see Cassidy et al., Reference Cassidy, Woodhouse, Cooper, Hoffman, Powell, Rodenberg, Berlin, Ziv, Amaya-Jackson and Greenberg2005; and Woodhouse & Cassidy, Reference Woodhouse and Cassidy2009, who note that providing physical contact until the infant is fully calmed may be a more powerful predictor of later security than the general sensitivity of the parent's response). Basic research examining the extent to which infant distress occurs in relation to the attachment behavioral system will provide an important foundation for further work.

Additional mediational pathways: Genetics, cognitions, and emotions

Following the discovery of the transmission gap, several studies examined the possibility of a genetic mediating mechanism. However, neither behavior–genetic nor molecular–genetic research so far indicates a genetic component to individual differences in secure vs. insecure attachment, although mixed findings have emerged concerning a genetic vulnerability for infant disorganized attachment (Bakermans-Kranenburg & van IJzendoorn, Reference Bakermans-Kranenburg and van IJzendoorn2004, Reference Bakermans-Kranenburg and van IJzendoorn2007; Bokhorst et al., Reference Bokhorst, Bakermans-Kranenburg, Fearon, van IJzendoorn, Fonagy and Schuengel2003; Fearon et al., Reference Fearon, van IJzendoorn, Fonagy, Bakermans-Kranenburg, Schengel and Bokhorst2006; Roisman & Fraley, Reference Roisman and Fraley2008). (For evidence that variability in infants' serotonin transporter linked polymorphic region predicts not whether infants are secure or insecure but their subtype of security or subtype of insecurity, see Raby et al., Reference Raby, Cicchetti, Carlson, Cutuli, Englund and Egeland2012.) Future research should examine other genes and Gene × Environment interactions (for examples from primate research, see Suomi, Reference Suomi2011).

Despite a conceptual model of intergenerational transmission in which maternal behavior is central, examination of additional linking mechanisms purported to underlie maternal behavior, such as maternal cognitions and emotions, will continue to be important. Perhaps such factors may be more reliably measured than maternal behavior, and if they are, mediating relations may emerge to shed light on mechanisms of transmission (e.g., Bernier & Dozier, Reference Bernier and Dozier2003). Moreover, from a clinical standpoint, factors thought to underlie maternal behavior may be more amenable targets of intervention than maternal behavior itself. For example, continued examination of maternal cognition through the study of constructs such as reflective functioning and maternal insightfulness may shed light on the link between mother and child attachment (Oppenheim & Koren-Karie, Reference Oppenheim, Koren-Karie and Zeanah2009; Slade, Sadler, & Mayes, Reference Slade, Sadler, Mayes, Berlin, Ziv, Amaya-Jackson and Greenberg2005). These constructs refer to the extent to which a mother can see the world from her infant's point of view while also considering her own mental state. There is evidence that these and other components of maternal cognition (e.g., perceptions of the baby, attributions about infant behavior and emotions, maternal mind-mindedness) are linked to maternal and/or child attachment, and additional research is needed to clarify the extent to which these components mediate the link between the two (e.g., Leerkes & Siepak, Reference Leerkes and Siepak2006; Zeanah, Benoit, Hirshberg, Barton, & Regan, Reference Zeanah, Benoit, Hirshberg, Barton and Regan1994).

Another aspect of maternal functioning that should prove fruitful for researchers examining the transmission gap is maternal emotion regulation. As Cassidy (Reference Cassidy2006) has proposed, much maternal insensitivity can be recast as a failure of maternal emotion regulation. That is, when mothers themselves become dysregulated in the face of child behavior or child emotions that they find distressing (particularly child distress), their maternal behavior is more likely to be driven by their own dysregulation rather than by the needs of the child (see also Slade, in press). Evidence that maternal emotion-regulation capacities contribute to problematic parenting and insecure attachment has been reported (Leerkes et al., Reference Leerkes, Parade and Gudmundson2011; Lorber & O'Leary, Reference Lorber and O'Leary2005), as have data indicating that maternal state of mind with respect to attachment (i.e., maternal secure base script knowledge) is uniquely related to maternal physiological regulation in response to infant cries (but not in response to infant laughter; Groh & Roisman, Reference Groh and Roisman2009). Unfortunately, although there is a sound conceptual and empirical basis for maternal emotion regulation as a mediator of the link between maternal and child attachment, there has been no empirical examination of this possibility.

In sum, the direction of future work depends on researchers' goals. If the goal is to understand the maternal behavior that mediates the link between maternal state of mind and child attachment, then the focus must be on maternal behavior. However, if the goal is to understand what factors may guide maternal behavior, and as such may themselves be successful targets of intervention, then examination of factors such as maternal cognitions and emotions should prove useful as well.

Caregiving as a function of adult attachment style

Although most researchers using self-report measures of adult attachment have not focused on links with parenting, there is a substantial and growing body of literature (more than 50 published studies) that addresses this link (for a review, see Jones, Cassidy, & Shaver, Reference Jones, Cassidy and Shaver2013). Whereas researchers using the AAI have focused mainly on links between adults' AAI classifications and their observed parenting behaviors, attachment style researchers have focused mainly on links between adult attachment style and self-reported parenting cognitions and emotions (reviewed by Mikulincer & Shaver, Reference Mikulincer and Shaver2007). However, the few studies in which self-report attachment measures were used to predict parenting behavior have found support for predicted associations (e.g., Edelstein et al., Reference Edelstein, Alexander, Shaver, Schaaf, Quas and Lovas2004; Mills-Koonce et al., Reference Mills-Koonce, Appleyard, Barnett, Deng, Putallaz and Cox2011; Rholes, Simpson, & Blakely, Reference Rholes, Simpson and Blakely1995, Study 1; Selcuk et al., Reference Selcuk, Günaydin, Sumer, Harma, Salman and Hazan2010). This is especially the case for maternal self-reported attachment-related avoidance. (Note that each of these studies was conducted with mothers only [Edelstein et al., included four fathers], so caution is warranted in generalizing these findings to fathers.)

It would be useful to have more studies of adult attachment styles and observed parenting behavior. It would also be important to conduct longitudinal and intergenerational research using self-report measures. Prospective research is needed examining the extent to which adult attachment styles predict both parenting behaviors and infant attachment (see Mayseless, Sharabany, & Sagi, Reference Mayseless, Sharabany and Sagi1997; and Volling, Notaro, & Larsen, Reference Volling, Notaro and Larsen1998, for mixed evidence concerning parents' adult attachment style as a predictor of infant attachment). Of related interest to researchers examining attachment styles and parenting will be longitudinal research examining the developmental precursors of adult attachment as measured with self-report measures. (See Fraley, Roisman, Booth-LaForce, Tresch Owen, & Holland, Reference Fraley, Roisman, Booth-LaForce, Tresch Owen and Holland2013; and Zayas, Mischel, Shoda, & Aber, Reference Zayas, Mischel, Shoda and Aber2011, for evidence that self-reported attachment style in adolescence and early adulthood is predictable from participants' mothers' behavior during the participants' infancy and early childhood.)

In general, we need more merging of social and developmental research traditions. It would be useful to include both the AAI and self-report attachment style measures in studies of parenting behaviors and cognitions. It would also be useful to know how the two kinds of measures relate similarly and differently to parenting variables. Scharf and Mayseless (Reference Scharf and Mayseless2011) included both kinds of measures and found that both of them prospectively predicted parenting cognitions (e.g., perceived ability to take care of children), and in some cases, the self-report measure yielded significant predictions when the AAI did not (e.g., desire to have children). From the viewpoint of making predictions for practical or applied purposes, it is beneficial that both interview and self-report measures predict important outcomes but sometimes do so in nonredundant ways, thus increasing the amount of explained variance.

Mothers and fathers

It is unclear whether it is best to think of a single kind of parental caregiving system in humans or of separate maternal and paternal caregiving systems. Harlow proposed separate maternal and paternal systems in primates (e.g., Harlow, Harlow, & Hansen, Reference Harlow, Harlow, Hansen and Rheingold1963). Within a modern evolutionary perspective, the existence of separate maternal and paternal caregiving systems is readily understood. Because mothers and fathers may differ substantially in the extent to which the survival of any one child enhances their overall fitness, their parenting behavior may differ. The inclusion of fathers in future attachment research is crucial. We contend that the field's continued focus on mothers is more likely to reflect the difficulty of recruiting fathers as research participants than a lack of interest in fathers. Bowlby, after all, was careful to use the term attachment figure rather than mother, because of his belief that, although biological mothers typically serve as principal attachment figures, other figures such as fathers, adoptive mothers, grandparents, and childcare providers can also serve as attachment figures. Researchers presume it is the nature of the interaction rather than the category of the individual that is important to the child. In addition, the inclusion of fathers will permit examination of attachment within a family-systems perspective (Byng-Hall, Reference Byng-Hall, Cassidy and Shaver1999; Johnson, Reference Johnson, Cassidy and Shaver2008). Future research should examine (a) whether the precursors of infant–father attachment are similar to or different from the precursors of infant–mother attachment, (b) whether the Strange Situation best captures the quality of infant–father attachments (some have suggested that it does not; Grossmann Grossman, Kindler, & Zimmermann, Reference Grossmann, Grossmann, Kindler, Zimmermann, Cassidy and Shaver2008), (c) the influence of infants’ relationships with fathers and father figures on their subsequent security and mental health, (d) possible differences in the working models children have of mothers and fathers, and (e) possible influences of parents' relationship with each other on the child's sense of having, or not having, a secure base (Bretherton, Reference Bretherton2010; Davies & Cummings, Reference Davies and Cummings1994).

Bowlby's theory of attachment and psychopathology

Bowlby used Waddington's (Reference Waddington1957) developmental pathways model to explain how early attachment relates to later developmental outcomes, including psychopathology. According to this model, developmental outcomes are a product of the interaction of early childhood experiences and current context (at any later age). Early attachment is not expected to be perfectly predictive of later outcomes. Moreover, attachment insecurity per se is not psychopathology nor does it guarantee pathological outcomes. Instead, insecurity in infancy and early childhood is thought to be a risk factor for later psychopathology if subsequent development occurs in the context of other risk factors (e.g., poverty, parental psychopathology, abuse). Security is a protective factor that may buffer against emotional problems when later risks are present (for a review, see Sroufe et al., Reference Sroufe, Carlson, Levy and Egeland1999).

Attachment and internalizing/externalizing behavior problems: State of the field

Over the past few decades, there have been many studies of early attachment and child mental health. The findings are complicated and difficult to summarize, as explained by Fearon, Bakermans-Kranenburg, van IJzendoorn, Lapsley, and Roisman (Reference Fearon, Bakermans-Kranenburg, van IJzendoorn, Lapsley and Roisman2010, p. 437): “With the sheer volume, range, and diversity of studies . . . it has become virtually impossible to provide a clear narrative account of the status of the evidence concerning this critical issue in developmental science” (italics added). Studies contributing to this body of work have used diverse samples and different methods and measures and have yielded inconsistent and, at times, contradictory results. Fortunately, two recent meta-analyses (Fearon et al., Reference Fearon, Bakermans-Kranenburg, van IJzendoorn, Lapsley and Roisman2010; Groh, Roisman, van IJzendoorn, Bakermans-Kranenburg, & Fearon, Reference Groh, Roisman, van IJzendoorn, Bakermans-Kranenburg and Fearon2012) provide quantitative estimates of the degree of association between child attachment and internalizing/externalizing symptoms.

The meta-analyses revealed that insecurity (avoidant, ambivalent, and disorganized combined) was related to higher rates of internalizing and externalizing symptoms (though the link appears to be stronger for externalizing symptoms). When the subtypes of insecurity were examined individually, avoidance and disorganization were each significantly related to higher rates of externalizing problems, but only avoidance was significantly related to internalizing problems. Ambivalence was not significantly related to internalizing or externalizing. Contrary to expectations, neither meta-analysis yielded much support for an interaction of child attachment and contextual risk in predicting behavior problems. For example, neither meta-analysis found support for the predicted child attachment by socioeconomic status (SES) interaction. However, given that high versus low SES is a rather imprecise measure of the numerous psychosocial risk factors that could contribute (individually and additively) to behavior problems, along with evidence from large sample studies supporting an attachment by risk interaction (e.g., Fearon & Belsky, Reference Fearon and Belsky2011), these results should be interpreted cautiously. In sum, the answer to the question “Is early attachment status related to later mental health difficulties?” is a resounding yes, but the precise nature of the connections remains unclear.

Attachment and psychopathology: Gaps in the research and future directions

More research is needed on mechanisms, or mediators, that help to explain how insecurity, or a particular form of insecure attachment, leads in some cases to psychopathology. These mechanisms should be considered at different levels of analysis: neurological, hormonal, cognitive, behavioral, and social–interactional. Mediators may include difficulties in emotion regulation and deficits in social skills, for example. Given the documented links between early attachment and emotion regulation and physiological stress responses (Cassidy, Reference Cassidy1994; Spangler & Grossmann, Reference Spangler and Grossmann1993), as well as the role of emotion dysregulation and HPA axis irregularities in psychopathology (Gunnar & Vazquez, Reference Gunnar, Vazquez, Cicchetti and Cohen2006; Kring & Sloan, Reference Kring and Sloan2010), emotion regulation seems to be a promising target for mechanism research. More research is also needed on potential moderators and risk factors, such as age, gender, personality, traumas and losses, SES, exposure to family and neighborhood violence. Researchers should consider the cumulative effects of multiple risk factors as well as interactions among risk factors (Belsky & Fearon, Reference Belsky and Fearon2002; Fearon & Belsky, Reference Fearon and Belsky2011; Kazdin & Kagan, Reference Kazdin and Kagan1994).

Given that most research on the mental health sequelae of early attachment has focused on internalizing and externalizing symptoms in nonclinical samples, future research should focus more on clinically significant problems and consider specific clinical disorders. The CBCL is not a measure of psychopathology, although it does indicate risk for eventual psychopathology (Koot & Verhulst, Reference Koot and Verhulst1992; Verhulst, Koot, & Van der Ende, Reference Verhulst, Koot and Van der Ende1994). Future research should address why the link between attachment and problematic behaviors is stronger for externalizing than for internalizing problems and whether this difference holds for diagnosable pathology (e.g., conduct disorder or major depression). This may be partially a measurement issue. The CBCL is often completed, with reference to a child, by a parent, a teacher, or both. It may be easier to see and remember externalizing behaviors than it is to notice whether a child is experiencing anxiety, sadness, or internal conflicts. Another important diagnostic issue is comorbidity. It is very common for clinicians to assign a person to multiple diagnostic categories. Perhaps attachment theory and related measures could help to identify common processes underlying comorbid conditions and suggest where their roots lie (Mineka, Watson, & Clark, Reference Mineka, Watson and Clark1998). One likely possibility is emotion regulation and dysregulation influenced by early experiences with parents.

Moreover, additional research is needed on the precise nature of the early childhood predictive factors and issues of causation. Is the issue really attachment status at age 1, for example, or is it continual insecure attachment across years of development? We also need to know whether attachment status per se is the issue or whether, for example, poor parenting predicts both attachment classification and psychopathology. Answering these questions will require studies using repeated assessments of attachment, parenting, context, and psychopathology. Further, there is increasing recognition of the importance of genetics and Gene × Environment interactions in understanding the development of psychopathology (e.g., Moffitt, Reference Moffitt2005). Given preliminary evidence for genetic influences on disorganized attachment (Lakatos et al., Reference Lakatos, Toth, Nemoda, Ney, Sasvari-Szekely and Gervai2000), as well as evidence for a Gene × Early Maternal Sensitivity interaction in predicting mental health outcomes (Bakermans-Kranenburg & van IJzendoorn, Reference Bakermans-Kranenburg and van IJzendoorn2006), this area of inquiry is a very promising avenue for further research. There is also growing evidence concerning environmental effects on gene expression (i.e., epigenetics; Meaney, Reference Meaney2010). Especially interesting is the possibility that secure attachment may protect a child from the expression of risky genotypes (see Kochanska, Philibert, & Barry, Reference Kochanska, Philibert and Barry2009, for preliminary evidence).

Interventions with infants and young children at risk for insecure attachment

Just as the study of individual differences in attachment began with the study of infants, so the systematic study of how to create attachment-related change focused initially on attempts to alter the developmental trajectory of infants who are at risk of developing or maintaining insecure attachments. Given the substantial and converging longitudinal research underscoring the risks associated with insecure attachment (e.g., poorer mental and physical health and lower social competence), the need to intervene to reduce the risk of insecure attachment is clear. In the past 20 years, researchers and clinicians have developed numerous therapeutic programs to prevent or reduce insecure attachment (Bakermans-Kranenburg, van IJzendoorn, & Juffer, Reference Bakermans-Kranenburg, van IJzendoorn and Juffer2003; Egeland, Weinfield, Bosquet, & Cheng, Reference Egeland, Weinfield, Bosquet, Cheng, Osofsky and Fitzgerald2000; see Berlin, Zeanah, & Lieberman, Reference Berlin, Zeanah, Lieberman, Cassidy and Shaver2008, for a review). Even though a handful of attachment interventions have shown initial success (e.g., Bernard et al., Reference Bernard, Dozier, Bick, Lewis-Morrarty, Lindhiem and Carlson2012; Cassidy et al., Reference Cassidy, Woodhouse, Sherman, Stupica and Lejuez2011; Cicchetti, Rogosch, & Toth, Reference Cicchetti, Rogosch and Toth2006; Hoffman, Marvin, Cooper, & Powell, Reference Hoffman, Marvin, Cooper and Powell2006; Klein Velderman, Bakermans-Kranenburg, Juffer, & van IJzendoorn, Reference Klein Velderman, Bakermans-Kranenburg, Juffer and van IJzendoorn2006; Lieberman, Weston, & Pawl, Reference Lieberman, Weston and Pawl1991; Lyons-Ruth & Easterbrooks, Reference Lyons-Ruth and Easterbrooks2006; Sadler et al., Reference Sadler, Slade, Close, Webb, Simpson and Fennie2013; Toth, Rogosch, Manly, & Cicchetti, Reference Toth, Rogosch, Manly and Cicchetti2006; van den Boom, Reference van den Boom1994, Reference van den Boom1995), we have as yet a collection of individual initiatives with little replication. More research is needed to identify the critical targets of attachment interventions and to understand the process of change.

Interventions during adolescence

Adolescence is a period of increased risk-taking behavior and mental health disorders (Centers for Disease Control and Prevention, 2012; Roberts, Attkisson, & Rosenblatt, Reference Roberts, Attkisson and Rosenblatt1998), and attempts to reduce these problematic symptoms and behaviors are important. The dramatic biological and cognitive changes that occur during adolescence have led some researchers to consider this period as a second sensitive period (e.g., Andersen & Teicher, Reference Andersen and Teicher2008; Guttmannova et al., Reference Guttmannova, Bailey, Hill, Lee, Hawkins and Woods2011), and as such, adolescents may be particularly open to environmental interventions that can lead to improved functioning.

Interventions designed to help adolescents are typically targeted at the reduction of specific problems (e.g., depression or delinquency), and there are many effective interventions for helping troubled adolescents (e.g., Dishion & Kavanagh, Reference Dishion and Kavanagh2003; Stein, Zitner, & Jensen, Reference Stein, Zitner and Jensen2006). Given evidence that the link between attachment security and healthy functioning continues to exist in adolescence (Dykas, Ziv, & Cassidy, Reference Dykas, Ziv and Cassidy2008; Kobak & Sceery, Reference Kobak and Sceery1988; see Allen, Reference Allen, Cassidy and Shaver2008, for a review), increasing adolescent attachment security may also be an effective means of reducing adolescent problems. In other words, does an intervention focused on the adolescent–parent attachment relationship increase the likelihood of reduced problematic adolescent behavior? There has been remarkably little research on this important question. Moreover, the associations among intervention, attachment, and psychopathology may be more complex in adolescence than they are in infancy and early childhood. Another important way to examine the connections among these factors is to design studies to determine whether reducing adolescent psychopathology contributes to secure adolescent attachment.

In an intervention designed to enhance adolescent–parent attachment security, with the idea being that increased security will reduce the prevalence of adolescent problems, questions about intervention approaches arise. For instance, should an intervention include the parent, the adolescent, or both? Bretherton (Reference Bretherton1992) described Bowlby as the first family therapist because of his proposition that change in a child's attachment to a parent is possible only when there is change within the interaction between the child and the parent (see also Byng-Hall, Reference Byng-Hall, Cassidy and Shaver1999). There has been almost no examination of this proposition in adolescence. One randomized controlled trial (Diamond et al., Reference Diamond, Wintersteen, Brown, Diamond, Gallop and Shelef2010) showed that attachment-based family therapy was more efficacious than enhanced usual care in reducing adolescents' depressive symptoms and suicidal ideation.