Temperament as a potential developmental influence on attachment

Infant differences in temperament may influence the develolbment of mother-infant attachment either directly or indirectly.

Temfierarnent as a Potential Deuelobmental Influence on AtLachment H . Hall Goldsmith, Donna L. Bradshaw Loretta A . Rieser-Danner

The developmental significance of temperament lies primarily in its impact on the social context of the child, according to most recent conceptualizations (Goldsmith and Camps, 1982; Lerner and Lemer, 1983; Thomas and Chess, 1977). In infancy, the central figure in the baby’s social world is the attachment figure, and thus a major issue for temperament research is: How do individual differences in the various dimensions of temperament influence the developing attachment relationship? The joint study of attachment and temperament may prove particularly instructive because it involves one variable that is primarily relational-attachmentWe appreciate the comments of Jacqueline V. Lerner, Richard M.h e r , Mary Rothbart, and James Connell on an earlier draft of this chapter. In addition, conversations with Wanda Bronson, Joseph Campos, Joan Stevenson-Hinde, Ross Thompson, and Everett Waters contributed directly to our thinking about issues of temperament and attachment. Goldsmith was supported by a Research Career Development Award (HD-00694)during the preparation of this chapter, and the Spencer Foundation funded Bradshaw and Goldsmith’s research reported herein. J. V. h a and R M. Lrma (Ed%).Tmprmmaf and Social Inkration During Infancy a d Childhmd New Dimdons tor Child Dcvelopmrns no. 31. San Frandsco:J w B a y . Mar& 19%.

5

and another that basically encompasses differences at the level of the individual-temperament We first analyze this issue conceptually and then proceed to the empirical studies. We do not propose that variations in the quality of attachment are wholly-or even largely-due to infants’ temperament differences. Simply correlating parents’ perceptions of temperament with Ainsworth’s Suange Situation Assessment of security of attachment (henceforth referred to as, simply, the strange situation) is only a crude first approach to the issue. At the outset, we must explain our understanding of the terms attachmentand temperament. By attachment, we refer to the widely known theory of Bowlby (1969, 1973, 1980),as popularized in the United States by Ainsworth (1969, 1972; Ainsworth and others, 1978) and others. Briefly, Sroufe and Waters (1977, p. 1185) define the attachment system as an “organizational construct.” “Attachment is not viewed as a static trait; rather, it has the status of an intervening variable or an organizational construct, to be evaluated in terms of its integrative power . . . attachment refers to an affective tie between infant and care-giver and to a behavioral system, flexibly operating in terms of set goals, mediated by feeling and in interaction with other behavioral systems. In this view, behavior is predictably influenced by context rather than constant across situations.” An extremely large proportion of attachment research employs Ainsworth and Wittig’s (1969) “strange situation” to assess the quality, or security, of attachment. This procedure entails two brief (three-minute) separations and subsequent reunions involving the infant, the mother, and a female stranger. While a mother’s behavior is supposed to follow a standard script, variability in infant behavior is rated on several interactive categories, the most crucial of which are avoidance, resistance, proximityseeking, and contact-maintenance during the two mother-child reunion episodes. In addition, infants are traditionally classified into one of eight subgroups within the major classifications of secure (type B), insecureavoidant (type A), or insecure-ambivalent (type C). Although the interactive category scores guide the classification of attachment security, weight is also given to the occurrence of specific signs of insecurity (Stroufe and Waters, 1977). Some of the most important of these are failure of the infant to greet the mother on her return, instances of angry resistance (for example, struggling when held by the mother), and inability of the mother to soothe the infant,. Quality of attachment thus assessed is intended to reflect the infant’s feelings of security and trust developed in the context of the relationship with the care-giver. Temperament, on h e other hand, is a rubric for several relatively independent dimensions of individual variability related to early personality. We have quite spedfic theoretical notions about the meaning of temperament (Goldsmith and Campos, 1982, 1985); however, in this chapter we typically use a conceptualization general enough to encompass the views of most current temperament theorists (Buss and Plomin, 1984; Chess and

7 Thomas, 1984; Goldsmith, 1985; Lerner and Lemer, 1983; Rothbart and Denyberry, 1981; Stevenson and Graham, 1982; Strelau, 1983). Examples of temperament traits that have been proposed by various investigators include activity, attention span, distractibility, disuess-proneness, fearfulness, persistence, reactivity, sociability, and soothability. Temperament dimensions are usually expected to be fairly stable and to have biological origins, with some qualifications. With regard to stability, the primary qualifications are that stabiIity should be evident within developmental periods, although not necessarily across periods of developmental transitions, or "biobehavioral shifts" (Emde and others, 1976; Goldsmith and Campos, 1982; Plomin, 1983; Rothbart, 1981). Even between periods of developmental reorganization, stability is not absolute but highly relative to other (often ill-specified) aspects of behavior. Stability is expected at the level of the latent construct rather than that of overt behavior, and stability in some temperament characteristics may be apparent only in the face of environmental suessors. We have particularly emphasized the emotional content of temperament characteristics (Goldsmith and Campos, 1982; 1985). Because attachment refers to an emotional bond, it is reasonable to investigate how the development of this bond may be mediated by individual differences in emotionality, or temperament. For instance, what input might the infant's or mother's characteristic degree of emotional expressiveness have on the interpersonal emotions involved in the attachment system-love, crust, or sense of security? What role will individual differences, on the parts of both members of the dyad, in sensitivity to the signals indicative of joy, fear, anger, interest, and other primary emotions have on the same interpersonal emotions? These are more specific components of the general question of the interrelation of temperament and attachment.

Temperament in Attachment Theory Both temperament and attachment are thought to be major organizers of socioemotional behavior. Both are also believed to predict later social outcomes, including behavior problems in school and sociability with peers (see Bates, forthcoming; Hinde and others, 1985; Keogh, Chapter Five of this volume; Sroufe, 1983; Lenerz and others, forthcoming). Despite this overlap in domains of influence, there is controversy over suggestions that temperament characteristics may in any significant manner determine the nature of individual differences in the quality of attachment (for example, see Sroufe, 1985). In discussing this issue, it is dseful to differentiate purely theoretical issues from the vagaries of the strange-situation procedures and parental report measures of temperament. In this section, we focus exclusively on the theoretical issues. The Limited Place of Tmpemment in Attachment Theory. Part of the reason the temperament-attachment issue can be contentious is that in neither area do contemporary theories accommodate the other construct

8

very well. However, the concept of temperament can be integrated with attachment theory in at least two ways, and with little distortion of Bowlby’s and Ainsworth’s approaches. Temperament dimensions can be understood as mediators of the social interactional processes underlying the formation and maintenance of the attachment bond (see Crockenberg, Chapter Three of this volume). In addition, temperament dimensions may make relatively direct contributions to individual differences in specific, observable attachment behaviors to key organizing influences in attachment theory, such as susceptibility to fear (see Thompson, Chapter Two of this volume). According to the ethological-organizationalapproach, factors that determine both the initial formation of the attachment bond and subsequent variation among dyads in its quality are (1) the sensitivity of the attachment figure in responding to the infant’s signals and (2) the amount and nature of interaction in the dyad. Interactions involving comforting when the infant is tired, ill, or frightened are deemed particularly important because the function of attachment behavior is the provision of protection and security to the infant (Bowlby, 1969). Patterns of mother-infant interaction-even in the first three months of life-predict patterns of attachment behavior at twelve months of age (Ainsworth and others, 1978). A corollary of this hypothesis regarding the origin of individual differences in attachment is that any variable that affects the nature of mother-infant interaction-especially involving distress and comfortingpotentially contributes to the quality of attachment. Historically, attachment researchers have focused on the variables of maternal sensitivity and responsiveness to the infant’s signals. However, several studies indicate that infant variations in temperament, among other infant characteristics, can influence the course of mother-infant interaction during the early months (see Crockenberg, Chapter Three of this volume). Consistent with this evidence, several researchers propose interactive models in speculating on the temperament’s contributions to the development of attachment (Bradshaw and others, 1984; Goldsmith and Camps, 1982; Weber and others, forthcoming; Rothbart and Derryberry, 1981; Bates, forthcoming). The Rothbart and Derryberry (1981, p. 68) position is representative: “As important as the mother’s sensitivity and flexibility may be, the role of the child‘s constitutional capadties and limitations in shaping her behavior should not be underestimated. Nor should the sensitivity and flexibility of the infant be neglected, for infants vary greatly in their capacity to augment or reduce their own reactivity, and to bring distress or pleasure to their care-givers. It seems essential that the mother-infant interaction and the resulting attachment process be viewed as a function of two intricate and flexible interactional systems, which can achieve a ‘balance’ in a number of ways.” Classical attachment theory adopts a different viewpoint. Although

9

both Bowlby and Ainsworth acknowledge that genetic or constitutional “biases of the infant” (Bowlby, 1969) may influence the mother, they both conclude that the mother’s characteristics primarily determine the quality of interaction and, therefore, of attachment (Ainsworth, 1979; Bowlby, 1969). In support, Ainsworth cites data from her longitudinal study of twenty-three mothers and infants in Baltimore; these data seem to indicate greater stability in maternal than in infant behavior during the first six months. There were also somewhat higher correlations between maternal behavior, in one quarter, and infant behavior (for example, frequency and duration of crying), in the next, than vice-versa (Bell and Ainsworth, 1972; Blehar and others, 1977). However, the de-emphasisof the infant variables may be unwarranted. Ainsworth based her conclusions regarding direction of effects on simple correlational analyses that are insufficient for resolving the issue. The cross-quarter correlations of infant behavior with subsequent maternal behavior were frequently only slightly lower than those of maternal behavior with later infant behavior. Thus, the question of whether infants’ temperament dimensions may influence attachment by influencing the quality of mother-infant interaction was not settled by this seminal study. Since then, evidence for maternal influences on security of attachment has continued to accumulate, as has, to a lesser degree, evidence for temperament-like influences. Establishing causal priority remains an elusive-even questionable-goal. Besides possibly influencing individual differences in attachment via its effect on mother-infant interaction, variations in temperament may affect the manifestation of certain attachment behaviors more directly. Theoretically, temperament differences might readily be incorporated into Bowlby’s control sytems model of the internal organization of attachment behavior in the infant. According to this model, attachment behavior is organized around a set goal, conceived of either as the physial degree of proximity to the attachment figure (Bowlby, 1969) or as the psychological degree of sense of security (Sroufeand Waters, 1977).Attachment behaviors such as proximity-seeking are manifested when the set goal is shifted so that the child’s current distance from the mother or current feelings of security are no longer within the required range, either as a consequence of external circumstances (alarming events) or of internal factors (increasing fatigue, hunger, or illness). Anxious attachment, which is manifested in part by overly clinging behavior, reflects an attachment system with a chronically low threshold for activation. In such cases, the slightest uncertainty or hint of potential stress results in deviation from the system’s set goal and consequently elicits attachment behaviors aimed at reduang this deviation (Bowlby, 1973). Thus, individual differences in anxiety versus security of attachment are considered by Bowlby to be closely related to variation in susceptibility to fear. In fact, according to Ainsworth and others (1978), fear is the feel-

10 ingstate that is opposite of a sense of security. Although Bowlby (1973) asserts that “Genetic differences play some part in accounting for variance between individuals in susceptibility to fear” (p. 187), he attributes the largest portion of the variance to the infant’s assessment of the attachment figure’s probable availability. This assessment, and therefore the child’s susceptibility to fear, is thought to derive from the infant’s actual experiences of the care-giver’s availability and responsiveness, and is believed to influence the chronic level of the set goal in the child’s attachment system (whereas transitory environmental events or internal changes such as hunger affect it only temporarily). In contrast, from the perspective of most theories of temperament, fearfulness-the tendency to react with distress and withdrawal in the face of novel situations-is an important aspect of temperament. Fear per se is a dimension in the approaches of Goldsmith and Campos (1982, 1985) and Rothbart (1981), and i t is a component of Buss and Plomin’s (1984) “emotionality,” Kagan, Reznick, Clarke, Snidman, and Garcia-Coll’s (1984) “inhibition,” and Thomas and Chess’s (1977) “mood” and “adaptability.” Theoretically, then, temperament variation in fearfulness, independent of the mother’s behavior toward the child, could influence the chronic levels of the set goal of the attachment, such that temperamentally fearful infants would tend to have lower thresholds for activation of attachment behaviors. Infants low in fearfulness would tend to have higher settings and be able to tolerate greater time and space apart from the attachment figure. It may be noted that another way to conceive of temperamental input into the developing attachment relationship is to ask how individual differences in the fear system might affect the “working model’’ of the mother. For instance, one aspect of the infant’s working model of the mother is the perception of her as a secure base for exploration of the environment. An infant whose fear system is chronically easily activated tends to experience correspondingly quick deactivation of the exploratory system; such an infant seldom moves far away from the mother. Less exploration means fewer opportunities to use the mother as a secure base. Hence, the secure-baseaspect of the working model is not often tested and is not highly elaborate. Beside fear, other temperament dimensions may influence the child’s internal working model of the attachment figure. For example, a difficult-to-soothe child may be more likely to experience the mother’s interventions as insensitive because only rarely will her efforts quickly succeed in overcoming distress. Despite prompt and frequent maternal responses, such infants may develop working models of the attachment figure as not well able to meet their needs, resulting in a tendency toward anxious attachment. One can also construct scenarios whereby infant differences in sociability, proneness to anger, and other temperament dimensions directly or

11 indirectly affect the evolving attachment system. Indeed, when we focus on the individual infant, the particular profile of temperament characteristics-rather than the infant’s standing on any one dimension-constitutes the temperament input into the attachment system. In summary, the roles of temperament in attachment variation may be integrated into the traditional ethological attachment framework in at least two ways: (1) as influences on the course of mother-infant interaction, which ultimately determines the child’s security of attachment, and (2) as more independent contributors to differences in the organization of the attachment system (determining the level of set goal of the attachment system). Undoubtedly, many attachment theorists would reply that although infants may have some effect on the quantity or quality of adultinfant interactions, or although infants may differ temperamentally in proneness to fear, the only important factor is how mothers respond to these differences-that infant differences have no direct effect on attachment security (see Waters and Deane, 1982). As we shall discuss later, the empirical data are presently inadequate to differentiate these alternative interpretations. Conceptions of Temperament in the Strange Situation In much attachment research, the A, B, and C classification derived from the strange situation are taken as synonymous with the concept of security of attachment. Thus,arguments about the role of temperament in the development of attachment-pro and con-center on the interpretation of strange-situation behavior. We believe that strange-situation assessment has been inappropriately reified as a measure of attachment, but we will not pursue this issue here (see Connell and Goldsmith, 1982; Campos and others, 1983). Attempts to Reintmjwet Strange-Situation Behavior in T m of T m p e r a e Only rarely, and perhaps heuristically, have investigators even seemed to suggest that temperament is a wholly sufficient alternative to security of attachment in explaining strange-situation assessment. With regard to behavior in the strange situation, Kagan (1982, p. 24) states, “A child who becomes distressed following maternal departure is more likely to rush to and greet the mother than one who fails to cry or one who is minimally distressed by the departure. As a result the former child is more likely to be classified as securely attached. Infants who ignoIr or do not greet the mother on her return, because they were not upset, are more likely to be classified as less securely attached. This is a aitical problem because there is good muon to believe that the child’s temperamental disposition to become distressed in uncertain situations makes an important contribution to behavior in the strange situation (Tennes and Carter, 1973; Tennes and others, 1977; Tennes and Vernadakis, 1977; Kagan and others,

12

1978)” Kagan concludes that classifications of security of attachment “are probably not totally independent of the child’s temperamental tendencies. Nonetheless, it is probable that the quality of the interactive relation between mother and infant does make a contribution.. . . However, because we are unable to rule out the likelihood of a serious interaction between the infant’s temperament and mother’s practices, the exact contribution of maternal behavior remains unclear” (p. 25, 26). In the context of an article mainly concerned with questioning the concept of neonatal bonding, Chess and Thomas (1982, p. 220) adopt perhaps a stronger position. They observe that “the items of the infant’s behavior in the Ainsworth Strange Situation could appropriatley be rated under the temperament categories of approachlwithdrawal, adaptability, quality of mood, and intensity” and that later assessments of competence could have similar interpretations of temperament. These positions have provoked counterarguments seeking to dismiss the possibility of direct and strong influences of temperament on the quality of attachment. Attem$ts to Dismiss T m w a m e n t as a Contributor to Attachment. As Bradshaw and others (1984), Goldsmith and Campos (1982), Rothbart and Denyberry (1981), Sroufe (1985), and others have pointed out, there are a number of ways that we can conceive of possible relations between temperament and attachment. The empirical arguments that have been raised attempt to counter suggestions that temperament may have a direct influence, rather than an interactive one, on the classification of security of attachment from the Strange Situation Assessment. One such argument, derived from the findings of Lamb (1978), Main and Weston (1981), and Grossman and others, (1981), is that some infants are classified differently when assessed in the strange situation with the mother versus the father, thus negating a temperament interpretation (Thompson and Lamb, 1984; Sroufe, 1985). The obvious underlying assumption is that temperament characteristics are consistent across the contexts of maternal and paternal interaction. Even granting this pointwhich many temperament researchers would not-there are less obvious, questionable assumptions in this argument. One of these assumptions is that temperament and maternal behavior are wholly alternative explanations for variation in attachment Classification; a temperament contribution of, say, one quarter of the variation is compatible with the typically observed degree of similarity in attachment to mother versus father. Another tacit assumption is that the quality of attachment to the father can be validly assessed using the A, B, and C system in the strange situation. The evidence linking attachment classification with prior caregiving patterns comes only from studies of mother-infant interaction. On the other hand, there is substantial evidence that, in traditional families, the father plays a qualitatively different role in child-rearing than does the mother (see Lamb, 1981). Therefore, it seems unlikely that a laboratory

13

procedure designed specifically to elicit signs of the security of the motherinfant relationship will serve the same function, or serve it to the same degree, for the father-infant relationship. Still another problem with the argument lies in its empirical backing. In the Main and Weston (1981) and Grossman and others (1981) studies, assessment of the attachment to the mother versus the father was separated by six months, opening the possibility that instability of attachment was partially responsible for the apparent parent attachment differences. The one study that assessed attachment to the mother and the father with only one week intervening did find a significant, but modest, association between the two classifications-and only when a B versus non-B contrast was examined (Lamb, 1978).The issue is a difficult one: Carryover effects make short intervals between testing inadvisible (Ainsworth and others, 1978),but long intervals confound parental gender differences with the substantial instability that is observed in many, but not all, studies (see Campos and others, 1983, for an empirical review of the stability evidence). A second major counterargument to temperament interpretations is that attachment is subject to change if the care-giver undergoes changes in life stress (Sroufe, 1985). There are really two parts to this argument; it seems clear that attachment classification can be unstable, but Egeland and Farber (1984) recently cast doubt on whether this instability is associated with life stress. Moreover, the assumptions necessary to validate this counterargument against temperament input into the Strange Situation Assessment can again be questioned. h e temperament dimensions themselves rigidly stable? Are temperament dimensions hypothesized to be the major determinants of attachment rather than one of several contributors? Is the “instability” in the Strange Situation Assessment real rather than largely reflective of measurement error? Other lines of argument against temperament determination rely on evidence that the Strange Situation Assessment predicts and is predicted by maternal behavior in other contexts (Sroufe, 1985). But is the influence of maternal sensitivity and responsiveness so strong that no other explanatory factors are needed? If previous studies have shown that security of attachment is largely predictable from maternal sensitivity and responsiveness, there would be little justification for examining temperament as a direct effect on security. A review of the important predictive studies, however, reveals that 30 percent is probably an overestimate of the variance in attachment classification accounted for by any measure of maternal sensitivity or responsiveness. Table 1 shows the magnitude of the significant effects reported in selected studies. The effect size estimates in Table 1 index the degree of departure from the null hypothesis, which itself represents an effect size of zero. Metric-free effect sizes, such as the r (squared) estimates in Table I, facilitate comparison among studies. Unfortunately, most reports include insuffi-

14

cient statistical information for calculation of exact effect sizes. Thus, Table 1 shows the minimum and maximum possible effect sizes for each study; it includes only results that attained conventional levels of statistical significance. Table 1. Strength of Maternal Influences on Attachment Classification: Selected Studies EffectSize study

Predictm

Ainsworth and others (1978) Blehar and others (1977)

First quarter home behaviors First quarter face-teface interaction Maternal responsivity Maternal sensitivity

Crockenberg Grossman and Grossman ( 1982)

Conversion to r2

Minimum Maximum Minimum Maximum d=.74

d=1.21

.12

.26

d=.82

d =1.38

-14

35

f =.lo

f =.25

.09

.20

d=.82

d=1.38

.12

.27

Note: The symbols d and f represent effect sizes derived from the reported analyses. The uansformation of these effect sizes to r values can be interpreted as proportion of variance explained by measures of maternal influences. Only relations that were strong enough to attain conventional leveIs of statistical significance ( 6 0 5 ) within the parameters of each study are included in this table.

Of course, the more general point that the attachment relationship is perturbed by noxious environmental influences does not depend on measures of maternal sensitivity. There are several other lines of evidence: insecure attachment among infants who have been abused or neglected and among those whose mothers perceive low social support and experience a great amount of stress (see Sroufe, 1985, for a brief review). Such evidence does not damage the notion that there is temperament input into the quality of attachment, and we are not concerned with the idea that putative attachment differences reflect nothing but temperament. By the same token, it should not be embarrassing to classical attachment theory that the typical effect sizes in Table 1 are so modest-human behavior is complex; but a significant portion of the variation, nevertheless, does remain to be explained. The degree of predictability observed in the strange-situation literature strongly implicates developmental factors other than maternal behavior. Temperamental influences are plausible candidates, but we should not expect them to account for a major portion of the variation in attachment. In effect, parallel analyses of the predictive power of infant behavior in the studies tabulated in Table 1 show effect sizes no larger than the tabulated values for maternal variables.

15

Temperament as a Predictor of Security of Attachment Versus Tmperament as a Predictor of Strange Situation Behavior That Is Not Central to Attachment. From the vantage point of attachment theory, the significant question regarding the relation of temperament to the strange situation is whether it predicts the security of attachment, that is, the probability of instances of anger, resistance, and avoidance on the one hand, and of getting the mother, seeking proximity, and becoming soothed, on the other. Yet the strange situation elicits a wide range of behaviors and indeed was designed to activate and de-activate the behavioral systems of exploration and fear or wariness as well as the attachment system. The characteristic level of activation of the former two systems is, in the view of temperament theorists, influenced by the temperament dimensions of activity level and fearfulness. Some would even consider them roughly synonymous with temperament characteristics. Consequently, it is entirely conceivable that temperament may infuence strange-situation behavior without affecting attachment security per se. Rather, temperament characteristics would affect the interplay of the three systems and thus the degree to which the attachment system is activated. For instance, crying when the mother leaves the room could be influenced by temperamental proneness to distress but be unrelated to security. A similar argument is offered by Sroufe (1985), who points out that crying is characteristic not only of infants classified as type C but also of many labeled as type B. That is, crying is not fully reliable as a diagnostic sign. If one views diagnosis (classification)as a more probablistic enterprise, however, it is quite clear that crying is associated with the type C versus B or A classifications (Ainsworth and others, 1978; Connel and GoIdsmith, 1982; Miyake and others, 1985). Another possibility is that differences in temperament might underlie behavioral variation within the broad B classification, or they might differentiate A and C varieties of anxious attachment, as suggested by Bronson (as cited in Sroufe, 1985). Another way to view the issue is to ask whether the strange situation contains adequate precautions to prevent temperment “contamination” of security measures. By virtue of their roles in fearfulness and exploration, temperament dimensions may contaminate the assessment of the quality of attachment. The valid assessment of the quality of attachment in the strange situation depends on the assumption that the context ought to activate the attachment system in every normal child and, furthermore, activate it to approximately the same degree (that is, the strange situation is argued to be moderately stressful for every child). The quality 6f the attachment system’s functioning obviously cannot be measured unless it is activated, and the degree of activation should not be confounded with the measure of quality. We raise the possibility that the strange situation does not strongly activate the attachment system of infants

16

who are highly sociable with strangers, active in exploration, and low in fearfulness, and that the measurement of attachment quality is therefore contaminated. Similar arguments, at the level of national groups, have been offered in interpreting cross-cultural differences in the relative frequencies of the A, B, and C types. Let us summarize the point in concrete terms: Readers experienced in scoring the strange situation will recall instances when tentative impressions based on behavior during the first reunion are disconfinned by quite clear signs of either secure or insecure attachment during the second reunion. In such’cases, the first separation episode apparently did not activate the attachment system sufficiently, but the cumulative stress added by the second separation did. If there were a third separation, might not the attachment classification of still more infants change? If so, one plausible interpretation is that temperament susceptibility to distress contaminates attachment classification. Emotional Reactions in the Strange Situation. That the strange situation stimulates a series of emotional reactions-in addition to the instrumental behaviors that primarily compose the Ainsworth scoring system-is clearly indicated by a number of studies (Connell, 1985; Connell and Goldsmith, 1982; Connell and Thompson, forthcoming; Gaensbauer and others, 1983; Waters and others, 1979). Using structural modeling approaches, Connell and his colleagues have shown that these emotional reactions predict Ainsworth’s interactive dimensions during the reunion episodes and that these predictions are replicable. That is, the infant’s separation distress, sociability with the stranger, and other emotional reactions within the strange situation guide, to some extent, subsequent interactions with the mother. It is likely that this guidance is mediated by the infant’s differing cognitive appraisals of the immediate context in differing emotional states (Ainsworth and others, 1978; Connell, 1985). Our view of temperament as individual differences in emotional expression becomes very relevant at this point. If emotional states during the strange situation predict social interactive behaviors, and if emotional traits (temperament)affect the probability of expression of corresponding states, then temperament affects the behavior thought to index attachment. However, finding correlations between emotional state and attachment behavior per se implies little about the relation of temperament to attachment. The important question concerns the external correlates of emotion in the strange situation. Are they predictable from expressed emotion within the context of the mother-infant relationship outside the strange situation? If so, this still does not necessarily implicate temperament as a root of strange-situationbehavior. However, the case for temperament is buttressed if those emotional reactions that predict the crucial reunion behaviors in the strange situation are predictable by infant emotion expressed outside the context of the mother-infant relationship. Of

17

course, an alternative explanation is also plausible-that the security of the developing attachment relationship leads to certain generalized emotional responses in contexts outside the key dyad. Sometimes the discussion of the temperament side of the temperament-attachment issue focuses, inappropriately, on only negative affectivity (or proneness to dismss, or “partial synonyms” such as difficulty). Activity, persistence, and sociability, as well as positive and negative emotionality, should be integrated into the discussion, and we review empirical work on these dimensions later in this chapter. A Role for Attachment in the Development of Temperament? While existing infant temperament may provide a context for the later development of the attachment relationship, the antecedents-on the maternal side-of attachment may also influence the further development of temperament. If true, would this fact hopelessly blur the issue? Our answer is “no,” not if the issue is to investigate processes of development rather than to maintain artificial boundaries between domains of study. We shall point out some possibilities for effects of attachment on temperament dimensions. Each of these possibilities must be understood in a context that recognizes (1) that temperament is malleable, (2) that aspects of the mother-infant relationship other than attachment may affect temperament, (3) that maternal precursors of attachment may affect temperament development before the attachment is fully functioning, and (4) that the issue is multidimensional in that attachment may differently influence activity, fearfulness, sociability, anger-proneness, persistence, and so forth. One possibility is that the attachment relationship serves a general shielding function, or fails to do so, with regard to the distress-related aspects of temperament. That is, the mother’s sensitive interaction moderates the stressful facets of the infant’s experience. Perhaps the most obvious idea is that a sense of felt security might reduce temperamental fearfulness. We are not referring to the phenomenon whereby an infant expresses less fear in the attachment figure’s presence than in other contexts. Rather, we refer to a process Bowlby (1969) suggested for older infants: The working model of a responsive mother associated with secure attachment allows the infant to become less fearful regardless of her presence, compared to an insecurely attached infant. Another notion is that certain maternal behavioral patterns-frustration of infant bids for contact-facilitate the development of angerproneness (angry responses in a number of contexts). However, if such maternal behavior leads independently to avoidant attachment (Main and others, 1985), it would not be the attachment relationship per se that influences anger-proneness. An idea inspired by Matas and others’ (1978) results is that securely attached infants become more persistent in tasks in the context of con-

18

smctive maternal interaction. Generalization of persistence could be considered a temperament change, depending on where one draws the boundary between temperament dimensions and related personality characteristics in the developing child Finally, if one agrees with Buss and Plomin (1984) that sociability is a primary temperament dimension, the research on peer behavior as a function of attachment history could be interpreted as another case of temperament change due to relationships. Methodological Problems in Studying the TemperamentAttachment Relationship Before turning to the empirical data, we shall point out that the joint study of temperament and attachment is complicated by more than the usual number of methodological problems. Most prominent is the very different nature of the entrenched measurement procedures: (1) Ainsworth and Wittig’s (1969) laboratory-based strange situation, a series of maternal separations and reunions designed to activate the attachment system; and (2)a number of care-giver report questionnaires used to infer the offspring’s standing on several temperament characteristics from covarying items (see Rothbart and Goldsmith, 1985, for a review). There are, however, more than historical reasons for the different methodologies in the two fields. Asymmetries in the Study of T m w a m e n t and Attachment Qualities of the Individual Versus Quality of the Dyad. Temperament is a set of qualities of the individual, and attachment is a construct that characterizes a relationship. Yet both are assessed from measures of the individual infant, a circumstance that opens at least the possibility of temperament contamination of security of attachment as measured in the strange situation. Although attachment researchers recognize the subtleties of maternal behavior during the strange situation can affect infant behavior, attachment security is judged almost exclusively on the basis of infant behavior (Ainsworthand others, 1978). The crucial infant behaviors in the strange situation are assumed to reflect the history of mother-infant interaction. Multidimenskmal Versus Categorical Characterization. As mentioned earlier, temperament is simply a convenient rubric for a class of several roughly orthogonal dimensions of infant behavior; their exact identity varies according to each theorist. For example, we have defined hmfierament as individual differences in the expression of the primary emotions and arousal (Goldsmith and Campos, 1982). From the vantage point of temperament phrasing the question as the relation of temperament to attachment is already too general. More productive questions concern, for

19

example, the relation of negative emotionality to attachment, or activity level t9 attachment Security of attachment, from the Bowlby and Ainsworth views, refers to a qualitative distinction between secure and insecure relationships. Security of attachment is not quantified in a dimensional sense, although interactive behaviors such as resistance and avoidance are. The implication of this particular asymmetry, besides the use of differerit statistical approaches to data analyses in the two areas, is that qualitative classification encourages typological thinking, whereas dimensional quantification fosters population thinking (Mayr, 1970). There is no necessary equation of dimensional approaches with traditional trait conceptualizations.That is, dimensional measures of attachment behavior do not necessarily carry traitlike connotation of temporal stability and cross-situational generality. The probable reason attachment is measured by a typological classification is that the research began in a clinical tradition, wherein diagnosis is a time-honoredfirst step. With occasional exceptions (Egeland and Farber, 1984), attachment researchers tend to write as though the strange situation classifies attachment security without error (except for scorers’ errors, of course). These measurement issues are discussed in detail by Connell and Goldsmith (1982). There are some exceptions to the distinction between temperament as a set of related dimensional constructs and attachment as a typological classification. Thomas and Chess (1977)have employed a typological characterization of “difficult,” “easy,” and “slow-to-warm-up”to describe the temperament of a portion of children, but recent research has employed the characterization less extensively. On the attachment side, dimensional approaches have surfaced. Connell (1985) and his colleagues (Connell and others, 1980) have used structural equation modeling to form best-fitting composites of interactive behavior ratings in both separation and reunion episodes. Although none of these composites is equated to security of attachment, they are an alternative to understanding the attachment-relatedbehavior that OCCUTS during the strange situation. In more traditional approaches, implicit dimensional thinking has also appeared. Main (1985)and Bates and others (1985)assign groups or subgroups to points on a “security of attachment” dimension. The combination of certain subgroups-for exmple, A with B1 and B4 with C (Hazen and Durrett, 1982) and B1 with B2 (Bretherton, 1985; Easterbrooks and Lamb, 1979; Thompson and Lamb, 1983)-implies the existence of an underlying continuum. Waters and his colleagues have moved completely away from the strange situation to construct a Q-sort measure of the security of attachment dimension from naturally occurring behavior (Waters and Dane, 1985; Waters and others, 1985). Using carefully devised behavioral descriptors, experienced attachment researchers performed criterion sorts that described the hypothetical securely attached child. Q-sorts that describe

actual children are correlated with the criterion, yielding a dimensional score, which we refer to later. In summary, the fundamental difference between measurement strategies in the temperament and attachment fields is beginning to fade. Degree to Which Attachment and Tmperament Must Both Be Invoked in Order to Control B e h a w . This is an asymmetry that is easily over-emphasized. In over-emphasizing the distinction, one may think of attachment as needing to be invoked before i t controls behavior, whereas temperament constructs may be construed as tending to exert control over behavior at all times. An outmoded view of traits might include the latter attribute, but modern trait concepts do not. Indeed, Campos and others (1983) view the emotions as organizational constructs, and it is individual differences in the emotions that are at the core of our concept of temperament (Goldsmith and Campos, 1982; 1985). Having cautioned h a t too much can be made of the issue, there is probably still some merit in viewing some temperament characteristics as providing a background against which attachment behavior can be invoked. Perhaps activity level influences a certain quality of motoric behavior across a great variety of contexts, whereas fearfulness must be invoked by rather specific stimuli, much as attachment is. But the distinction between background factor and invoked system is only a matter of degree. For example, Bretherton (1985) has argued that it is more fruitful to consider the attachment system as continuously operating-at different intensities-than as sometimes activated, sometimes not.

Overlap in the Timing of Development of Tmperament and Attachment

-if

It would be convenient if temperament dimensions developed very early during the first year of an infant’s life and the attachment relationship had its antecedents only late in the first year. Although most dimensions of temperament certainly have some degree of developmental priority over attachment, it is wishful thinking to suppose that the two kinds of constructs can be separated on a chronological basis alone. There is substantial evidence that infant-to-mother attachment becomes organized (a term that obscures our vast ignorance) during the second six months, but it has earlier antecedents (Ainsworth and others, 1978; Grossman and others, 1985). At six months attachment behaviors are already observable in many infants (see Bowlby, 1969). The timing of temperament development varies for different dimensions. The evidence comes from many sources: temperament-relatedbehaviors first appear at various periods during infancy (Emde and others, 1976; Izard and others, 1980); individual differences in various dimensions become stable at different times (Rothbart, 1981); and heritability appar-

21

ently varies-and tends to increase-during the early years (Goldsmith, 1983; Matheny, 1983; Torgersen and Kringlen, 1978).As is also the case for attachment (Ainsworth, 1967; Bowlby, 1969). the behaviors that express temperament undergo ontogenetic change. In the light of these and the other considerationswe have discussed, it is quite inappropriate to suggest that establishing a causal role for temperament in determining attachment differences requires establishing stability of temperament from the early weeks of life and concordance for attachment of identical twins reared apart-a sample virtually impossible to obtain (Sroufe, 1985). Empirical Studies of Temperament and Attachment in the Strange Situation t3wmi.m. A stmightforward first approach to the temperament-attachment question is to mrrelate the most typical measures of the two constructs. Although at the time of this writing few full reports have appeared in print, at least fifteen such analyses have been done. There are two factors that distinguish the enmes in this emerging literature: whether temperament is assessed prior to or concurrently with attachment and whether questionnaires or some other methods are used to measure temperament. Inevitably, our account of this literature is preliminary, and the meta-analyses needed to draw firm conclusions from the small-samplestudies with their modest effect sizes must await fuller publication of results. Before beginning our review, a comment concerning the aforementioned modest effect sizes seems appropriate. The notions of systems theory, multidirectional effects, organism-environmentaal transactions, and related ideas permeate the field of social development. Although these concepts tend to obfuscate some issues when they are not given concrete meaning, what scientist doubts how great the true complexity of behavioral development is? When developmental causes are numerous and their effects on some outcomes are interdependent, expectations regarding the predictive power of any one cause for the outcome variable must be adjusted-downward1 Despite the compelling nature of this conclusion, developmentalists are confronted with an entrenched statistical methodology, primarily identified with the analysis of variance, that is inaccurately translated to mean, “only big effects are important.” We shall try to avoid this mentality that implies that small pieces of the puzzle are unimportant-especially when there may simply be no large pieces. Unfortunately, the reader will not find all, or even most, of the hypotheses that we suggest to be tested in the existing literature. Not surprisingly for a recent area of investigation, the analyses are largely descriptive. R& of Tmpt?rrantent-Attachmt Studies. This review will discuss three themes. The first is that resistant behavior in the strange situa-

22 tion is associated with distress-proneness or negative affectivity in many other situations. The second, less well-supported, theme is that infants who show avoidance of the mother in the strange situation and are sometimes classified as insecurely attached (type A) tend to be perceived as possesing hedonically positive temperament characteristics. The third theme-which can be stated with some confidence-is that the first two themes rest on shaky empirical grounds and have not always been supported. Furthermore, questions pertaining to direction of effects largely remain to be addressed. There is a fourth empirical theme that is so obvious it will not be explicitly treated: Although temperament and security of attachment may be interrelated, they are by no means interchangable constructs. Table 2 shows some characteristicsof relevant studies. At the outset, it should be mentioned that all of the studies listed in Table 2 examine variables other than temperament and attachment or include special groups of subjects of one type or another. We have extracted the temperament-attachment results and have sometimes given them more emphasis than the original authors did, particularly when such treatment highlights the consistency of modest relations across studies. Bradshaw and others (1984) studied strange-situation behavior and maternal report of temperament via Rothbart’s (1981) Infant Behavior Questionnaire (IBQ). The crux of the relations found between the IBQ and strange-situationinteractive behaviors was that a broad positive versus negative affect factor derived from the IBQ predicted avoidant behavior (r=.42;) the more positive the affect, the more avoidant the infant in the strange situation, and resistance, proximity-seeking, and contact-maintaining (r = -24, -28, and -29, respectively). Viewing the results in terms of specific IBQ scales, the Duration of Orienting (persistence)Scale predicted avoidance (r = .48), whereas the Fear and Smiling and Laughter scales were associated with the interactive ratings of resistance, proximity-seeking, and contact-maintaining in the strange situation at a somewhat lower level. Thus, results suggest that infants who tend to avoid the mother in the strange situation are perceived as having their attention frequently engaged with objects in the home environment. This casts their avoidance behavior in a somewhat more positive light by emphasizing their capacity for sustaining attention without adult intervention, although it cannot be ruled out that lack of care-giver responsiveness is the source of this apparent independence. Although limited by the absence of infants classified as type C, this study’s other implication is that distress and inhibition directed toward a wide variety of sudden or novel stimuli, as indexed by the IBQ Fear Scale, relate to resistance toward the mother during reunion. Perhaps this latter relation is mediated by the tendency of more fearful infants to become distressed during separation episodes. The Rieser-Danner and others’ (1984) sample is very small, but one

pattern arises clearly from the data: Correlated temperament scales induding positive mood, the tendency to approach (versus withdraw) when confronted with new situations, and low-intensity responses predicted type A (four infants) versus B1, B2 versus B3 and B4 classifications. This result, similar in meaning to that of Bradshaw and others (1984), was achieved after deleting from the temperament scales items that depressed internal consistency estimates. The same pattern of results remained when the sample was expanded (Rieser-Danner, 1985).

Table 2. Description of Temperament-Attachment Studies ~~~

Authors (dnte)

N

Tmperammt Assessment (age in months)

Bradshaw and others (1984) Rieser-Danner and others (1984) Waters and others (1980) Egeland and Farber (1984)

40

Infant Behavior Questionnaire (1 BQ) (12)

23

Toddler Tmperament Questionnaire ( T T S ) (12)

Frodi and Sheldon (1982) Owen and Chase Lansdale (1982) Bates and others (1985) Crossman and others (1985) Gockenberg (1981) Weber and others (forthcoming) Meyer ( 1985) Belsky and others (1984) Miyake and others (1985) Stevenson-Hinde (1985) Thompson and Lamb (1984b)

100

Neonatal &havioral Assessment Scale (NBAS X 2)

189212

Neonatal: NBAS, nurses’ ratings Ratings during feeding (3.6)and play (6) Infant Temperament Questionnaire (ITQ) (6) Revised ITQ (RITQ) (a. 12)

40 132 63

Maternal and paternal I T S (12 & 1%)

49

Infant Charactcsistics Questionnaire RITQ and Maternal Perceptions Questionnaire Examiner ratings during Bayley examination Neonatal: NBAS X 3

48

Neonatal: NBAS X 2

36

Dimensions of Temperament Scale (DOTS) (13)

36