Patterns of Attachment: A Psychological Study of the Strange Situation - Mary D. Salter Ainsworth 2015
The Effects of Repetition of the Strange Situation
Measures and Methods of Assessment
Introduction
As Masters and Wellman (1974) have pointed out, laboratory studies to date have not demonstrated any significant degree of stability of attachment behaviors from one time to another or from one situation to another. Because of the significant degree of relationship shown by Sample 1 between behavior at home and behavior in the strange situation (see Chapter 7), it seemed unlikely to us that the kinds of behaviors highlighted by our strange-situation procedure could be grossly unreliable. Nevertheless, to select a way in which such stability could be tested, presented considerable difficulty. None of the conventional methods used to demonstrate stability of test performance seemed entirely satisfactory. Attachment behavior—and indeed behaviors antithetical to it, such as exploratory, avoidant, and resistant behavior—are demonstrably “situational.” That is, they are activated to different degrees of intensity in different situations. Furthermore, at low levels of intensity of activation, one set of attachment behaviors is likely to be manifest, whereas at high levels of activation another set is more likely. Therefore, any search for stability (at least as simply conceived) that assumed interchangeability of episodes, much as odd and even items of a multiitem test are considered interchangeable, seemed foredoomed to failure. No comparable laboratory situation had been devised that could serve as an “alternate scale” in the same way that two forms of an intelligence test are considered to be alternatives.
The only conventional method that held any hope of being appropriate for our problem was the test—retest method of assessing the stability of the behaviors at issue, and it is not entirely satisfactory. Three drawbacks are evident. First, an unfamiliar situation is not unfamiliar when it is encountered a second time. If a short time elapses between “test” and “retest” it will surely be recognized the second time, and once recognized, there will be anticipation of what is scheduled to happen later, so that behavior, even in the early episodes, may be affected by what previously happened in the later episodes. This might be especially likely to happen in a situation, such as this, in which behaviors in the later episodes are activated at high intensity. Second, if one allows a long time to elapse between “test” and “retest,” in the hope that memory of the first session will have faded, developmental changes may have taken place. If indeed such changes have occurred, one is faced with the problem of assessing stability in terms of “transformations”—that is, to assess the proposition that behavior at one age has “continuity” with behavior at a later age, despite possible substantial differences in the form of the behavior. Third, to the extent that one alters the situation enough that it would not be recognized when encountered a second time, one risks changing it enough that the situations, and hence the behaviors displayed therein, are not comparable from one session to another.
Nevertheless, we decided to undertake an assessment of stability using a test—retest method. We used an interval of 2 weeks. Later in the chapter we report two studies that used an interval of 6 months—the first by one of us (EW) that used a sample quite independent of our total sample of 106 infants, and the other by Connell (1976).
Effects
Let us first consider the test—retest study designed for the present investigation. The 23 infants of Sample 3 were introduced twice to the strange situation, once at 50 weeks and again at 52 weeks. The procedure was necessarily the same. The experimental room was also the same. The stranger and the toys were different but, we believed, equivalent. We expected, nevertheless, that 1-year-olds would recognize the situation after a mere 2 weeks lapse of time (we did not want to make it longer, lest developmental changes obscure the findings); and it was anticipated that the first session would affect behavior in the second session. Specifically, we expected that infants who had been distressed by separation in the first session would remember their distress and, anticipating separation-to-come in the second session, would be distressed even earlier in the second session, and more intensely distressed in comparable episodes. On the other hand, it seemed reasonable to expect that the infants who had experienced the first session without obvious distress might feel even more comfortable in the second session. Nevertheless it was expected that the major behaviors displayed in interaction with the mother—proximity/contact-seeking, contact-maintaining, avoidant, and resistant behaviors—would be correlated across the two sessions. Infants relatively high on one of these variables in the first session would also tend to be relatively high in the second session.
TABLE 21 Comparison of Meansa of Strange-Situation Measures for Sessions 1 and 2 and Correlations Between Sessions
A comparison between the two sessions was made in terms of these four behaviors, scored for the stranger as well as for the mother. The following behaviors were also compared: search behavior, crying, smiling at mother and at stranger, vocalizing to mother and to stranger, and touching the mother. Instead of dealing with each episode separately, scores for each of the two preseparation episodes were combined, as were scores for each of the two reunion episodes and, where relevant, scores for the separation episodes. It is clear from an examination of Table 21 that an infant’s behavior during the second session was affected by his experience during the first session. The following behaviors showed a significant increase from Session 1 to Session 2: seeking proximity and contact with the mother in both preseparation and reunion episodes, maintaining contact with the mother in the reunion episodes, search for her in the separation episodes, and crying both in the separation episodes and in the situation as a whole. Thus both attachment behavior and distress were more intense in the second session.
Resistant behavior toward the mother did not increase significantly in the second session. Avoidance of the mother in the reunion episodes decreased significantly in the second session. Resistance to the stranger in the separation episodes increased in the second session. This increase seems to be linked to intensified separation distress; more babies cried during the second session and they cried more often; consequently the stranger intervened more in an effort to give comfort, offering more occasion for resistance to be manifested.
These findings are not the result of large changes in a few children and of little or no change in most of the others. An examination of individual cases showed that for 21 of the 23 infants, attachment behaviors were stronger in the second session than in the first. Thus our initial expectations were not wholly borne out. Infants who had manifested little or no distress during the first session did not feel more comfortable in the second; rather they, like those who had been distressed during the first session, were more anxious in the second. This suggests that our gross behavioral assessment of distress in Session 1, in terms of crying, was inadequate to identify all infants who experienced disturbed affect. Indeed, as reported in Chapter 13, Sroufe and Waters (1977) observed heart-rate acceleration in those strange-situation circumstances expected to introduce some stress both among infants who cried and among those who did not.
Two other investigations, both discussed in Chapter 9, included two strange-situation sessions within 2 weeks of each other, one session with the mother as the accompanying adult and the other with the father, with counterbalancing for order. Willemsen, Flaherty, Heaton, and Ritchey (1974) reported significant increases from Session 1 to Session 2 in proximity/contact seeking, contact maintaining, and crying, especially in the separation episodes and in Episode 3. This is congruent with our findings. (They did not assess avoidant or resistant behavior, however.) Lamb (in preparation) did not use our measures of proximity/contact seeking or contact maintaining, but found no order effects for crying or for any of his measures that reflect response to parent figures. It is possible that differences in measures account for the fact that his findings in regard to Session 1 vs. Session 2 differ from ours and from those of Willemsen and associates, although the fact that crying did not significantly increase in Session 2 tends to belie that explanation.
Table 21 also presents evidence of substantial stability from Session 1 to Session 2 in regard to interaction with the mother. A significant positive correlation was found for contact-maintaining behavior with the mother in the reunion episodes. Babies who sought to maintain contact during Session 1 also tended to do so, although more strongly, during Session 2. Those babies who sought proximity or contact with their mothers in the preseparation episodes of Session 1 also tended to do so, but more strongly, in Session 2. In addition, some babies who had not sought proximity during the first session did so during the second—a fact that makes the correlation coefficient of .56 seem remarkably high. Proximity and contact seeking in the reunion episodes did not emerge as stable, however. Not only did babies who had shown little or no proximity seeking in the first session show more in the second, but some babies who received high scores in the first session were so upset in the second that their proximity seeking became less active, and they resorted to signaling behavior, which received lower scores.
Conspicuous in Table 21 is the correlation for avoiding the mother. Although the mean avoidance score for Session 2 was lower than that for Session 1, the correlation between the two sessions is substantial. Babies who avoided their mothers during the reunion episodes of Session 1 tended to do so also in Session 2, although during Session 2 their avoidance behavior was weaker and more intermingled with attachment behavior. Resistant behavior directed toward the mother was, however, essentially uncorrelated from Session 1 to Session 2.
The amount of crying during the two sessions was highly correlated even though there was significantly more crying during Session 2. Furthermore, infants who searched strongly during the separation episodes of Session 1 also tended to do so in Session 2. Resisting the stranger during the separation episodes also emerged as fairly stable and, as we suggested earlier, may be viewed as another measure that reflects separation distress.
Smiling, vocalizing, and touching are labeled as “discrete” behaviors in Table 21, to distinguish them from our scaled measures of interactive behavior. In light of the review by Masters and Wellman (1974), we did not expect these to show stability from Session 1 to Session 2, and as may be seen from Table 21, five of the six correlations, although positive, were too low to be significant. The one exception was a substantial positive correlation for smiling to the stranger in the separation episodes. (Maccoby and Feldman, 1972, who likewise used frequency measures of discrete behaviors, also found that behavior directed to the mother showed no substantial stability across ages, but that some behaviors directed toward the stranger did.) Furthermore, although five of these behavioral measures were higher in Session 2 than in Session 1, none of them was significantly different from one session to another.
In summary, the first strange-situation session seemed to sensitize the infants of this sample to separation from their mothers, so that they were more distressed and anxious in the second session than in the first. Nevertheless, certain of the behaviors examined were reasonably stable from the first to the second session. It was shown in Chapter 6 that most of these behaviors—specifically, avoidance in the reunion episodes, proximity seeking in the preseparation episodes, contact maintaining in the reunion episodes, and resistance to the stranger in the separation episodes—are significant in discriminating the classificatory groups, A, B, and C, from one another. The only highly discriminating behavior that was not stable from Session 1 to Session 2 was resistance to the mother in the reunion episodes, a behavior characteristic of Group C. Only two babies in Sample 3 were sufficiently resistant during Session 1 to be classified in Group C; therefore the apparent lack of stability in this behavior may be due merely to the relatively small amount of resistant behavior shown by infants of this particular sample. In general, the correlations are remarkably high when viewed as test—retest coefficients for behavioral measures in a situational test, for such tests are notorious for yielding low test—retest reliability (Block, 1972).
Infants were also classified independently according to their patterns of behavior in each of Sessions 1 and 2. The teams of judges who classified behavior in one session were “blind” as to the behavior in the other session. The unexpected outcome is shown in Table 22. It may be seen that none of the seven infants classified in Group A on the basis of their behavior in Session 1 was so classified in Session 2; instead they were classified in Group B in Session 2—that is, with a 0% hit rate. Of the 14 infants classified in Group B in Session 1, 12 were so classified in Session 2; two had moved to Group C—an 86% hit rate. Of the two infants classified in Group C in Session 1, one was classified in Group B in Session 2—a 50% hit rate. Overall the stability of classification was 57%. This does not, however, suggest a random pattern of change. The Group-A babies tended to move into Subgroups B1 or B2 in Session 2, the B1 babies to B2, and the B2 babies B3. There was only one Group-C infant who was an exception to the trend toward more “normative” classification in Session 2.1
TABLE 22 Stability of Strange-Situation Classification at 12 Months When Compared with Classifications Obtained from a Second Testinga
From this and the direction of the differences between the means of the interactive-behavioral scores themselves, it is clear that the effect of the repetition of the strange situation after a brief 2-week interval was to increase distress and the intensity of attachment behavior, while at the same time the “negative” behaviors of avoidance and resistance decreased or remained at about the same level of intensity. Nevertheless, the substantial correlations between sessions for some of the scores of interactive behavior toward the mother, and for crying both in the separation episodes and overall, suggest substantial stability of individual differences across the span of two weeks. On the other hand, the shifts in classification suggest, at least at first glance, that patternings of behavior are not stable. The Group-A infants yield the most important evidence relevant to this issue. In Session 2, Group-A babies were still avoidant, but their attachment behavior in the reunion episodes had intensified to an extent that judges perforce paid more attention to proximity/contact seeking and contact maintaining than to avoidance when making classifications. This does not suggest so much that individual differences were unstable as that the classificatory system did not provide for temporary stress-induced shifts in behavioral patterning. There are thus two issues: the short-term effects of separation in an unfamiliar situation, and the long-term stability of individual differences in the organization of the attachment relationship. Let us defer a discussion of these issues, however, until after examining Connell’s and Waters’ findings.
Connell (1976) followed up 49 infants of his second sample (see Chapter 9) from 12 months of age until they were 18 months old, at which time he administered the strange situation a second time. The means of interactive-behavioral measures for each of the two sessions are shown in Table 23. It may be seen that for most of the measures the means at 12 and 18 months are very similar. Although Connell did not cite the significance of the differences between pairs of means, it is perhaps worth noting that the average child at 18 months sought proximity to and contact with his mother (in Episodes 3 and 8) more strongly than at 12 months. The average child at 12 months showed both more avoidance of and resistance to the stranger at 18 months than at 12 months, and also manifested more contact-maintaining behavior to the stranger. The only noteworthy decrease from 12 to 18 months was in resistance to the mother in Episode 8. It seems likely that these differences reflect developmental changes. The 18-month-old shows somewhat stronger attachment behavior than the 12-month-old and somewhat more disturbed behavior toward a stranger. On the whole, however, the changes were not great, and one gains the impression that 18-month-old toddlers behave similarly to 12-month-old infants in the strange situation.
Of perhaps more interest is that such a large proportion of the measures were significantly correlated between 12 and 18 months. This was especially the case for measures of interactive behavior directed toward the mother; only proximity/contact seeking and contact maintaining in preseparation Episode, 3 showed no correlation between the two sessions. There were also significant correlations for interactive behaviors directed toward the stranger, except for contact maintaining in the separation episodes and proximity/contact seeking in Episode 4. In short, individual differences in behavior toward the mother in the reunion episodes showed a substantial degree of stability, whereas behavior toward the stranger also showed some stability, especially resistance and avoidance in the second separation episode.
TABLE 23 Comparisons of Means of and Correlation Between Strange-Situation Measures at 12 and 18 Monthsa
Behavioral Measure |
Episode |
Session at 12 mos. |
Session at 18 mos. |
r |
|
||||
Interactive Behavior to M |
||||
Contact maintaining |
3 |
2.3 |
2.8 |
.06 |
5 |
2.9 |
2.9 |
.55c |
|
8 |
4.3 |
3.6 |
.57c |
|
Proximity/contact seeking |
3 |
2.8 |
3.8 |
.04 |
5 |
3.8 |
3.8 |
.62c |
|
8 |
4.0 |
4.8 |
.45c |
|
Resistance |
5 |
1.3 |
1.5 |
.51c |
8 |
2.4 |
1.3 |
.53c |
|
Avoidance |
5 |
2.5 |
2.0 |
.33b |
8 |
2.3 |
1.8 |
.52c |
|
Interactive Behavior to S |
||||
Contact maintaining |
4 |
1.6 |
1.3 |
00 |
7 |
1.5 |
2.7 |
.04 |
|
Proximity/contact seeking |
4 |
1.3 |
1.6 |
−.06 |
7 |
1.7 |
2.6 |
.37c |
|
Resistance |
4 |
1.8 |
2.4 |
.34b |
7 |
2.4 |
3.5 |
.64c |
|
Avoidance |
4 |
2.7 |
2.8 |
.37b |
7 |
2.0 |
3.6 |
.62c |
a Data from Connell (1976).
b p < .05.
a p < .01.
Of major interest, however, is the remarkable degree of consistency that Connell found between the classifications of his infants at 12 months and those at 18 months. For his classifications he used his “multivariate classifier” (see Chapter 9). The results are included in Table 22. It may be seen that 38 of the 47 children—or 80.9%—were classified into A-B-C groups as before. Thus the stability of patterning of strange-situation behavior emerged as even greater than the stability of the individual scores that entered into the patterning.
Waters (1978) also used the strange situation at 12 and 18 months with a sample of 50 infants. He also correlated the interactive-behavioral measures in Sessions 1 and 2 to ascertain their stability over the intervening 6 months. These findings, including those for search behavior, are shown in Table 24. It may be seen that the correlations for behavior directed toward the mother are positive and significant, except for distance interaction in the preseparation episodes, and search in the separation episodes. The highest correlation was found for contact-maintaining behavior in the preseparation episodes (r = .720), a finding that perhaps reflects the fact that Group-C infants are discriminated from both A and B babies in terms of this behavior. Substantial correlations were also found for avoidant and resistant behavior in the reunion episodes—behaviors that had previously been found (Chapter 6) important in the discrimination among groups. To a lesser extent proximity/contact seeking in the preseparation episodes was also fairly stable.
Four of 10 correlations for behavior toward the stranger reached the .05 level of significance. All of these were low, and all seem related to the presence or absence of separation distress, for all occurred in the separation episodes—proximity/contact seeking, contact maintaining, resistance, and distance interaction. It is plain that behaviors directed toward the mother show more stability over time than behaviors directed toward the stranger.
TABLE 24 Correlations Between Measures at the 12-and 18-Month Testings for Interactive Behaviorsa
Pearson Correlation Coefficients Between Sessions 1 and 2 |
||||
Behavior to Mother |
Behavior to Stranger |
|||
Interactive Measure |
Episodes |
r |
Episodes |
r |
|
||||
Contact maintaining |
2 & 3 |
.720c |
3 |
−.020 |
5 & 8 |
.300b |
4 & 7 |
.320b |
|
Proximity/contact seeking |
2 & 3 |
.423c |
3 |
.033 |
5 & 8 |
.303b |
4 & 7 |
.286b |
|
Avoidance |
5 & 8 |
.621c |
3 |
−.207 |
4 & 7 |
.229 |
|||
Resistance |
5 & 8 |
.508c |
3 |
−.056 |
4 & 7 |
.274b |
|||
Distance interaction |
2 & 3 |
.065 |
3 |
.180 |
5 & 8 |
.308b |
4 & 7 |
.319b |
|
Search |
4, 6, & 7 |
.147 |
a Data from Waters (1978).
b p < .05.
c p < .01.
Waters was particularly interested in comparing the stability of our measures of interactive behavior—which he calls “categorical scores”—with those of “discrete” behaviors based on frequency measures and considered independently of one another rather than in combination and/or as alternatives as in the interactive measures. The correlations based on discrete measures, for 30 subjects randomly chosen from his total of 50, are shown in Table 25.
In contrast to the categorical scores of interactive behavior toward the mother, shown in Tables 21, 23, and 24, it may be seen that these discrete measures tended to have no stability from one session to another. Exceptions were vocalizing to the mother and touching her in the preseparation episodes. As for the behavior toward the stranger, there was no apparent stability across time. Waters’ measure of crying, like ours, was also a frequency measure and is not included in Table 25. It showed some indication of stability. Crying (including fussing) in the separation episodes (r = .397), reunion episodes (r = .416), and all episodes (r = .394) were significantly correlated (p < .05) between 12 and 18 months.
TABLE 25 Correlations Between Measures at the 12- and 18-Month Testings for Discrete Behaviorsa
Pearson Correlation Coefficients Between Sessions 1 and 2 |
||||
Behavior to Mother |
Behavior to Stranger |
|||
Discrete-Behavioral Measures |
Episodes |
r |
Episodes |
r |
|
||||
Looking at |
2 & 3 |
.072 |
3 |
−.047 |
5 & 8 |
.224 |
4 & 7 |
.109 |
|
Vocalizing to |
2 & 3 |
.357b |
3 |
.119 |
5 & 8 |
−.073 |
4 & 7 |
.240 |
|
Smiling at |
2 & 3 |
.143 |
3 |
−.165 |
5 & 8 |
−.048 |
4 & 7 |
.194 |
|
Gesturing to |
2 & 3 |
−.116 |
3 |
−.083 |
5 & 8 |
−.107 |
4 & 7 |
−.103 |
|
Approaching |
2 & 3 |
−.151 |
3 |
.106 |
5 & 8 |
.043 |
4 & 7 |
.085 |
|
Touching |
2 & 3 |
.435b |
3 |
—c |
5 & 8 |
.105 |
4 & 7 |
.255 |
|
Baby holding onto |
2 & 3 |
—c |
3 |
—c |
5 & 8 |
.263 |
4 & 7 |
−.032 |
a Data from Waters (1978); N = 30.
a p < .05.
c Mean = 0 at one age level.
In striking contrast to his findings for the “discrete” measures, Waters obtained a very high degree of consistency between the classifications at 12 and 18 months. (See Table 22.) Using our standard classificatory instructions, he found an even more precise match between the two sets of classifications than Connell did in his sample. Of 10 Group-A infants at 12 months, nine were still so classified at 18 months, and the same was true of the 10 Group-C infants at 12 months. All of the 30 Group-B infants at 12 months were still so classified at 18 months. Forty-eight of the 50 infants—96%—were classified as before. Although Table 22 shows the results only for the main groups, Waters’ match even held quite well for subgroups.
In summary, both Waters and Connell found a remarkable degree of stability of A-B-C classification between 12 and 18 months, even though they used different methods of arriving at the classifications. The degree of stability over a 2-week period appeared less in our own data, but there was nevertheless a striking trend toward normative Group B in the second session—a trend that belies any argument that there were random shifts in classification attributable to “error.” In addition, all three studies showed a substantial correlation between Sessions 1 and 2 in regard to interactive behaviors directed toward the mother, and somewhat less stability in interactive behaviors directed toward the stranger. In particular all studies yielded evidence of stability for contact maintaining and avoidance in the reunion episodes, and Connell and Waters also found stability for proximity/contact seeking and resistance in the reunion episodes. The evidence of stability for the separate interactive scores was very much less striking than the evidence for the stability of classifications in Connell’s and Waters’ studies. As for the stability of discrete-behavioral measures, neither Waters nor we found as much evidence as for the stability of interactive-behavioral scores. Both studies found vocalizing to the mother to be marginally stable—Waters for the preseparation episodes, we for the reunion episodes. Waters obtained a significant cross-session correlation for touching mother in the preseparation episodes, and we for smiling to the stranger in the separation episodes. Neither of these sets of findings, separately or together, can be described as offering adequate evidence of stability.
Leaving the lack of stability of the discrete measures out of consideration, there still remains a remarkable paradox in that there is extremely high stability of the classifications (in Waters’ and Connell’s data) and yet evidence of only moderate levels of 12-to-18-month stability of the interactive-behavioral measures on which the A-B-C classifications were based. Especially because the stability of classifications was demonstrated in samples independent of ours in laboratories other than the one in which we developed our scoring system, we cannot question Connell’s and Waters’ findings. There are some changes in patterns of interactive behavior, some of them perhaps attributable to developmental changes; and yet there emerges this remarkable consistency in classifications. How are we to interpret this paradox?
One possibility is that the A-B-C classifications are so broad that they are insensitive to all but the most extreme changes in actual behavior. Under such circumstances it is not surprising to find stability of classification, and indeed such findings might be characterized as exaggerating the stability of the infant—mother attachment relationship during the period from 12 to 18 months. This seems unlikely in view of the shifts of classification when the intersession period lasts only 2 weeks—shifts that clearly are not attributable to random error. Another possibility is that the shifts during the period of 12 to 18 months nevertheless leave the patterning of the interactive behaviors more or less intact, and thus the A-B-C classifications remain relatively stable. In any event it appears that the patterning of interactive behaviors as represented in the classifications are stable in contrast with repeated failures to find stability in a variety of “discrete” variables, studied independently of one another (e.g., not only in the findings reported in this chapter but also in findings reported by Coates, Anderson, & Hartup, 1972b; Maccoby & Feldman, 1972; and as reviewed by Masters & Wellman, 1974).
To clarify the paradoxical nature of his findings, Waters (1978) performed an auxiliary analysis of his 12- to-18-month data to test the hypothesis that the A-B-C classifications are relatively insensitive to random variation in the interactive behaviors. Each interactive variable (e.g., contact maintaining, proximity/contact seeking, avoidance, etc.) scored at 12 months was randomly and independently transformed to produce a set of scores that reproduced the correlation between the 12- and 18-month data. The new data thus reproduced were in effect error analogues of the 18-month data. Data from our 105 12-month-olds and his 50 12-month-olds were used to develop classification equations that were then applied to the error data and the 18-month data. If the A-B-C classifications were insensitive to random variation in interactive behavior, the patterns of predicted classification should be the same in both sets of data. In fact, the “real” data and the error data produced strikingly different patterns of classification. In the “real” data, Group-B classifications were highly stable, and changes in the A and C classifications reliably drifted toward B classifications. In the error data, Group-B classifications often changed (46% vs. 13% in the real data), producing 16% A classifications and 30% C classifications from the 12-month Group-B data. In addition, when A and C classifications changed (48% and 45% respectively vs. 33% and 36% in the real data), the predicted classification was equally likely to be Group B or the other group (C or A as the case might be). The hypothesis that random variation in the interactive behaviors could underlie the stability of the A-B-C classifications was thus not confirmed.
Discussion
The correlation data in our study, as well as in Waters’ and Connell’s, show a substantial degree of stability over time in regard to the measures of interactive behavior toward the mother in the strange situation—both behavior in interaction with her when she is present and behavior reflecting distress when she is absent. On the other hand, there is somewhat less stability in regard to measures of interactive behavior toward the stranger. These findings are the reverse of those reported by Maccoby and Feldman (1972) (see Chapter 10), who found little stability in behavior directed toward the mother but more stability in response to the stranger. The discrepancies between their findings and those reported in this chapter may be attributable to either one or both of two factors: (1) that there is more stability between 12 and 18 months than between 24 and 30 or 36 months—namely, that relevant developmental changes are greater in the third than in the second year of life; and (2) that their measures were less suited than ours to the task of uncovering underlying continuities in behavior. Although it seems likely to us that both factors have played a part in making for discrepancies, here we wish to focus on differences in measures.
Except for their measures of “proximity” and “speak-smile-show,” Maccoby and Feldman used discrete, frequency measures. With the possible exception of crying, their frequency measures tended not to be stable between age levels and not to relate to other variables. In regard to this lack of stability of frequency measures, both our findings and Waters’ tend to confirm theirs. Furthermore, Maccoby and Feldman did not score proximity in the reunion episodes, and although they kept an eye open for discrete behaviors that might be related to proximity/contact seeking, contact maintaining, avoidance, and resistance (and indeed tallied them), they apparently found no evidence of stability across age. Waters’ examination of the stability of approach and touching in the reunion episodes bears out their impression of lack of stability when these behaviors are represented in separate frequency measures. The very measures that we find most stable across time (as well as most discriminating among individuals)—namely, the “categorical” measures of interactive behavior to the mother, especially in the reunion episodes—were plainly not adequately represented in the Maccoby and Feldman study. Coates, Anderson, and Hartup (1972b) not only used laboratory procedures that differed from those of the strange situation, but also focused on discrete measures. They also reported little stability over time. Because they relied chiefly on these two studies, it is perhaps not surprising that a conclusion of Masters and Wellman’s (1974) review was that attachment and attachment behavior lack stability over time.
Let us discuss the differences between discrete behavioral measures such as Maccoby and Feldman and Coates and associates used and our interactive-behavioral measures. There are two main differences. First, our interactive behavioral measures cover a variety of separate behavioral components, considering them interchangeable to some extent. Thus, for example, in the reunion episodes an infant may make a bid for increased proximity to his mother through active approach (perhaps making contact with her on his own initiative), or by gesture, or by other signals, (such as crying), or by some combination of these. Waters examined the stability of these separate behaviors from 12 to 18 months and found that none of them, except for crying, was significantly correlated across time. Both Waters and Connell, however, found evidence of stability of our proximity/contact-seeking measure. By 12 months of age, normal infants are capable of goal-corrected behavior, and this implies ability to use alternate means to achieve a set-goal, or to implement a plan. It seemed to us that the appropriate measures would be those that assessed the degree of initiative in achieving the set-goal and the degree of strength implicit in the behavior or sequence of behaviors adopted. On the other hand, there seems no reason to suppose that a baby will always use the same means to his ends. Indeed within one session a baby who actively approached in one reunion episode might merely reach and cry in another, although in both episodes he sought closer proximity to his mother. Presumably similar variations of means toward ends might well occur from one session to another. Frequency measures of component proximity-promoting behaviors, considered separately, do not take into account the fact that these behaviors are to some extent interchangeable.
Second, let us consider proximity-promoting behaviors, such as vocalizing and smiling. Maccoby and Feldman differentiated between these “distal” behaviors and “proximal” behaviors, such as those involved in active proximity seeking and contact maintaining, hypothesizing that a developmental shift from proximal to distal modes might occur over time, and that those who scored high in proximal behavior at one age point might score comparably high in distal behavior at another age point—implying that the “strength” of attachment thus remained constant. (Their data did not support the hypothesis.) A more useful distinction between the two sets of behaviors is in terms of the intensity of activation of the attachment-behavioral system. Vocalizing, smiling, and the like are most likely to occur under conditions of low intensity of activation of attachment behavior, and thus differ from the more active and/or intense proximity-seeking and contact-maintaining behaviors that tend to occur under conditions of high intensity of activation. Consequently the former are not increased in incidence by the conditions implicit in the strange-situation procedure that were intended to activate attachment behavior at high levels of intensity.
Furthermore, at low levels of activation of the attachment system attachment behaviors of all kinds, including vocalizing and smiling, are of relatively infrequent incidence. It will perhaps be recalled that in the “free-play” episode, Episode 2, infants smiled only once at the mother, on the average, and the mother’s noninterventive role undoubtedly offered little special instigation to vocalizing or smiling. Thus in regard to these low-intensity behaviors, laboratory studies encounter the problem of inadequate behavioral sampling. The 6 minutes of the two preseparation episodes together constitute too brief a time to yield an adequate sample. Thus for some of the discrete behaviors, stability between sessions cannot be adequately assessed because of the sampling problem. Even considering these to be interchangeable in some way, as in our categorical measure of distance interaction or in Maccoby and Feldman’s speak-smile-show measure, did not result in stability across time.
Of more concern to us than the stability of behavioral measures over time is the issue of the stability of the patterning of behaviors that is reflected in the classificatory system. Our hypothesis is that qualitative differences in the child—mother relationship, which are reflected in the patternings of strange-situation behavior and which in turn are reflected in the A-B-C classifications, tend to be stable over time. To be sure, we can conceive of shifts in the security/anxiety dimension of this relationship that are attributable either to substantial changes in the pattern of interaction between mother and child or to gross changes that influence that interaction (such as severe separation experiences) or to both in combination. Nevertheless our longitudinal study of mother—infant interaction throughout the first year of life (see Chapters 7 and 8) has led us to expect continuity to be more frequent than discontinuity. Certainly within a period as short as two weeks, discontinuities could be expected only in a very few special cases in which circumstances had intervened grossly to change the quality of the child—mother relationship. Over a period of six months, however, it might be expected that more such changes might occur.
Therefore, it is paradoxical that Waters and Connell found a high degree of stability in the nature of the child—mother attachment relationship over a period of six months, whereas we found less apparent stability over a period of two weeks, when the criterion of stability was a match in regard to strange-situation classification from two sessions separated by a lapse of time. When the two sessions are separated by only two weeks, there is clear-cut evidence that anxiety engendered by the first session carries over to influence behavior in the second. We interpret the findings to suggest that the anxiety pertains to the issue of the mother’s accessibility, and we believe that this issue is more or less restricted to the particular circumstances of the strange situation itself. In any event babies are more disturbed when they encounter a second strange situation shortly after they encountered the first, and this disturbance tends to break down the distinctions implicit in the classificatory system.
On the other hand, if the second session is some months after the first (as in Waters’ and Connell’s studies), the disturbance attributable to the first session seems to have dissipated, so that qualitative differences in the child—mother attachment relationship emerge as highly stable. To be sure, some of the few instances of instability may have been due to error in assessment or to chance variability in strange-situation behavior, but some may conceivably have come about because the relationship between child and mother had indeed undergone a qualitative change. With both of these possibilities open, it is indeed extraordinary that both Connell and Waters independently found such a high degree of stability over a 6-month period.
In conclusion, it appears that the individual differences that are stable over time are differences in the qualitative nature of the child’s attachment relationship to his mother. These are reflected in our system of classifying the patterns of his behavior in the strange situation. The components of the patterning are tapped chiefly by the categorical measures of interactive behavior, not all of which are attachment behaviors, and those behaviors that are most closely related to the patterning as well as being most stable over time occur under conditions of activation of the attachment system at high levels of intensity. This means that the mother-directed behaviors are best observed in the reunion episodes, for only in the case of a relatively few infants did the preseparation episodes offer high-intensity activation. Discrete behavioral measures neither relate to the patterning of behaviors that reflect quality of attachment nor tap stable separate aspects of behavior in themselves. As Tracy, Lamb, and Ainsworth (1976) found for approach behavior observed at home, it is not only the environmental context of approach behavior that enables one to judge whether it serves attachment or some other behavioral system, but also the behavioral context in which approach appears. Discrete frequency measures, by their very nature, ignore behavioral context. The categorical measures of interactive behavior, in contrast, take both environmental and behavioral context into account. Of all our methods of assessment, however, it is the classificatory system, including the subgroups, that through its focus on patterns of behavior gives most scope for considering behavioral context. We consider the issue of classification into subgroups in the next chapter.
Notes
1 Lamb (personal communication) also reported some slippage between A and B groups, but no consistent direction of slippage such as ours. He found the greatest instability in his small group of C babies. His investigation (in preparation) was intended, however, to compare the attachment relationships of the infant to mother and father. The fact that the attachment figure in Session 2 differed from that in Session 1 introduces a variable that tends to obscure any effect attributable to session.