Assessment of Maternal Interaction Style as a Precursor to Attachment

Maternal interaction style (often conceived as sensitivity) and security of attachment have long been considered to have an important relationship (Ainsworth et al., 1978; Grossmann et al., 1985; Belsky Rovine, and Taylor,1984; Isabella and Belsky,1991). However, the construct of maternal sensitivity is not consistently defined or measured in the literature. Using videotaped data of 33 mother-infant dyads, we identifed the relevant components of maternal sensitivity as related to attachment · outcomes. Data consisted of ( 1) six weekly naturalistic observations of free-play interaction in the home at 6 month and again at 9 months of age (i.e. 12 assessments for each dyad); (2) face-to-face interactions conducted in the laboratory at 6 and 9 months; (3) Ainsworth Strange Situation at 12 months during a laboratory visit; and ( 4) a Qsort measure of attachment security. Scoring systems appropriate to each of these assessments were used. Mul tiple home assessments were used so that a series of observations could be aggregated to form reliable measures of the maternal sensitivity scales. The effects of age level, setting, and number of observations considered were examined in terms of relation to attachment classification and attachment security measures. Results indicated that (1) aggregation of multiple home observations produces highly reliable and consistent assessment measures (2) laboratory measures are related to home observations (3) neither home nor laboratory observations predict attachment classifications outcomes and ( 4) home, but not laboratory observations, are related to security of attachment as measured by the Q-sort . Current findings are discussed in the context of previous attachment research. Methodological as well as theoretical explanations are considered to account for lack of relatedness between maternal sensitivity and attachment outcomes in the current study.

Maternal interaction style (often conceived as sensitivity) and security of attachment have long been considered to have an important relationship (Ainsworth et al., 1978;Grossmann et al., 1985;Belsky Rovine, and Taylor,1984;Isabella and Belsky,1991). However, the construct of maternal sensitivity is not consistently defined or measured in the literature.
Using videotaped data of 33 mother-infant dyads, we identifed the relevant components of maternal sensitivity as related to attachment · outcomes. Data consisted of ( 1) six weekly naturalistic observations of free-play interaction in the home at 6 month and again at 9 months of age (i.e. 12 assessments for each dyad); (2) face-to-face interactions conducted in the laboratory at 6 and 9 months; (3) Ainsworth Strange Situation at 12 months during a laboratory visit; and ( 4) a Qsort measure of attachment security.
Scoring systems appropriate to each of these assessments were used.
Mul tiple home assessments were used so that a series of observations could be aggregated to form reliable measures of the maternal sensitivity scales.
The effects of age level, setting, and number of observations considered were examined in terms of relation to attachment classification and attachment security measures.
Results indicated that (1) aggregation of multiple home observations produces highly reliable and consistent assessment measures (2) laboratory measures are related to home observations (3) neither home nor laboratory observations predict attachment classifications outcomes and ( 4) home, but not laboratory observations, are related to security of attachment as measured by the Q-sort .
Current findings are discussed in the context of previous attachment research.
Methodological as well as theoretical explanations are considered to account for lack of relatedness between maternal sensitivity and attachment outcomes in the current study.  Tables  Table 1  Table 2  Table 3  Table 4 lV Research ,,  Table 5 45 Table 6 46 Table 7 47 Table 8-A   48   Table 8-B   49  Table 9 50 Appendix I 5 1 Appendix II 52 Bibliography 53 Introduction Statement of the problem Maternal parenting behavior, characterized as sensitive has been proposed as an important contributor to mother-child attachment.
Sensitive parenting has also been associated with beneficial developmental, social and cognitive outcomes for the child , while insensitive parenting has been linked with insecure ,----------att~hment and less favora~le developmental outcomes. Debate continues, however over the proper operationalization of the construct of maternal sensitivity as well as its assessment.
The present study was designed to address some of the methodo o.gis:_ al~ssues in assessing th~ q_ uality of parenting as related to attachment. Most studies to date assess the mother -infant interaction at one or two distinct points during the course of the first year of life, and relate that qualitative measure to the attachment classification at twelve months. It is difficult to establish, on the basis of such investigations, how representative the discrete point of measurement is of the ongoing relationship of the dyad in question.
How much does an unusual visit by an experimenter affect the interaction? How peculiar or ordinary was that pa_ rticular day in the life of the family? More importantly, does the style of interactiol!, evolve or dramatically change over time and if so, how different ~ould the findings be if a different age level was chosen for assessment?
Attachment theory presumes accumulated expenence on the part of the child (Ainsworth, 1978;Bowlby, 1969). When has there been enough accumulation to justify proper assessment? Belsky, Rovine & Taylor (1984 ), for example found similar patterns relating maternal style to types of attachment at 6 and 9 months, but only the 9 month results reached significance. Other researchers have found significant relationships with assessments conducted as early as 3 months and 6 months of age (Egeland & Farber, 1984;Price, 1983Isabella & Belsky , 1991. Grossmann et al. (1985) suggested that repeated visits to the participants' homes would allow the observers to witness a greater variety of relevant behaviors as well as develop a more accurate perception of usual interactive patterns.
There is a lack of clarity, however in terms of the optimal period as well as the number of such assessments that may be necessary to gather a representative sample of interaction style.
® r '---The present study directly addressed this methodolo ical issue by co nd. ucting a _series of_ hoil}~ o~s~rv~_tions around the_ child's 6th and-.9.th-month of age. Through these repeated measures three major questions were addressed: (1) Is one time period _( _ §_ vs. -~ _r nonths) relatively more stable or representa~i~<:_ -in assessing the mother- Since the classifications are based on the child 's behaviors and not the mother's, the implicit assumption remains that assessment of the child's behavior in a heightened situation, will reflect and underscore the existing ongoing relationship, as experienced by the dyad. Attribution of a classification, then, does not speak to the components of the relationship or its precursors. The classification of attachment on the basis of the Strange Situation paradigm has become an accepted convention. The construct of attachment along with the operationally defined attachment behaviors as observed m the Strange Situation have been used almost exclusively to index the quality of attachment of a young child to his mother (Pederson et al., 1989).

@:-)
Recently, however, an alternative appro ach has been suggested and used by some researchers (Waters & Deane, 1985;Pederson et al., 1989;Vaughn and Waters,1990;Moran et al., in press The completed sort is then compared with the "criterion sort" (a sort of the prototypically secure child, as judged by a series of experts). The resulting correlation is interpreted as a continuous measure of the child's relative security of attachment.

Replication Studies of Maternal Sensitivity and Later Attachment
Several studies have provided support to the hypothetical association between type of mothering and attachment classification.
A close replication of the original Ainsworth study was reported by Grossmann et al. (1985) in a sample of longitudinally followed mother-child dyads in Northern Germany. Using global ratings of sensitivity, developed by Ainsworth et al, the authors found the infants of sensitive mothers more likely to be independently classified as securely attached than infants whose mothers were less sensitive over the course of the first year.

Maternal Attributes and Attachment Classifications
Some researchers have further explored the relative differences in style and behaviors of mothers of secure vs. anxiousavoidant vs. anxious-anxious-resistant babies.
It has been suggested that maternal style of interaction, falling on a continuum according to level of stimulation can serve as a differentiating measure (Belsky, Rovine, & Taylor, 1984). These authors found mothers of securely attached infants to demonstrate an "intermediate" level of interaction, as compared to the "overstimulation" and "neglect" which characterized the interaction style of mothe rs of anxious-avoidant and anxious-resistant infants, respectively. Egeland and Farber ( 1984) describe stable differences among these three sets of mothers in slightly different terms. Caretaking abilities , including general knowledge, timing, and responsivity as well as maternal feelings and attributions about motherhood were considered. As expected, mothers of securely attached babies, were most appropriately responsive, while mothers of anxious-avoidant babies were characterized as "indifferent" and "unavailable". Mothers of anx10usresistant babies suffered more from lack of awareness, than lack of interest, but also failed to provide proper, sensitive care.

Correlates of Insensitivity
In examining the extremes of parenting inadequacies , maltreated infants, described as suffering from "caretaking casualty" (Sameroff & Chandler, 1975), have been found to be more likely than their normative counterparts to demonstrate insecure attachment (Schneider-Rosen, et. al., 1985) and specifically to be classified as anxious-avoidant (Lyons-Ruth et al., 1987). Maltreating mothers were judged as interfering and "covertly hostile" when observed m the home, just prior to the strange situation assessment.
Among the numerous factors studied as predisposing of mothers to a relative quality of parenting have been age , (Ragozin et al. 1982), and depression (Radke-Yarrow et al, 1985;Field, 1988;Cohn et al. 1986). Teenage mothers, without proper intervention, have been shown to be at greater risk for displaying insensitive parenting styles leading to increased risk of infant developmental delay. (Field, 1980;Levine , et al., 1983 ). Maternal depression has been associated with disturbed face-to face interactions as well as insecure infant attachment (Cohn, et al., 1986).

Correlates of Sensitivity
"Positive" parenting styles have been associated with an array of beneficial outcomes. Cognitive development in both normative and delayed populations has been positively linked with sensitive parenting (Donovan & Leavitt, 1978;Mahoney, et al., 1985;Bakeman & Brown, 1980). Beckwith at al. (1976) found maternal sensitivity, m terms of responsiveness and appropriate stimulation, to be related to higher developmental scores for premature infants. Donovan & Leavitt (1978) and Mahoney et al., (1985) and Bornstein and Tarnis-LeMonda (1988) report similar findings in normative populations.

Operationalizing Sensitivity
In light of this evidence, a closer look at the construct of sensitivity is in order. Noted researchers in the field have emphasized various aspects of the maternal repertoire in their definitions of maternal sensitivity. According ly, several terms have been used interchangeably in the literature to refer to the underlying construct of optimal parenting.
As described above, Ainsworth's concept of sensitivity IS the appropriate and contingent responsiveness exhibited by the mother to her infant's cues. Ainsworth has relied primarily on naturalistic observations particularly in caretaking situations to arrive at global ratings of relative maternal sensitivity. (Ainsworth & Bell, 1969). Stem (1974) has focused on the timing and structure of the motherinfant interaction particularly during moments of social play.
A ttunemen t is the desired state of mutual responsiveness which IS attained by an infant and his mother, provided that the mother is able to perceive the infant's cues and adjust her behaviors to the appropriate level of stimulation. Such interactions, studied in detail during face-to-face interactions, are characterized by periods of mutual greeting, engagement and breaks. The sensitivity of the mother in this case would be most closely associated with her ability to tune up or down according to her infant's needs. Insensitive interaction is often characterized by intrusive, or overstimulating behaviors at times when the infant is sending signals for a break , or lack of interesting action when the infant is engaged and clearly available.
Detailed investigation of the face-to-face interaction has bee n the focus of study of Tronick, Als, & Brazelton (1980) Cohn, et al. (1986 and Kaye and Fogel (1980 (1972) reported that in their longitudinal sample, infants whose mothers responded to crying quickly and consistently in early infancy, had children who cried less, and used alternative communication more than children of mothers who systematically did not respond in this fashion.
Narrowing the construct even further some researchers have used responsivity alone as an implicit or explicit measure of sensitivity (Crockenberg & McCluskey, 1985;Lewis & Feiring (1989).
These researchers either distributed self-report questionnaires aimed at assessing responsivity patterns or actually counted frequency of responses on a time sampled basis. In interpreting their results these authors often equate these variables with measures of sensitivity.
By examining recent literature exploring the link between maternal sensitivity and attachment, the variability of conceptualization, as well as design and interpretation of findings becomes evident. In addition, behaviors of interest vary from free-play, feeding, caretaking, or "regular activities" chosen by the mother.

Method Subjects
Thirty-three mother-child dyads participated rn the study.
These subjects were participating in a larger study (N=50)  informed, however, that a minimum of ten minutes was to be observed for each of the three types of behaviors during the visit.
Total length of the home visit was usually between thirty and ninety minutes.
Mothers were also asked to complete a series of personality questionnaires at the onset of the study, as well as weekly questionnaires descriptive of the child's behavior.

Length of Study:
Home visits began when the infant was 3-4 months old and continued until 12-14 months of age.
Laboratory Procedures: Mothers made three lab visits with their infants at 6, 9, and 12 months of age. All procedures took place in a standard play room and were videotaped through a window in an adjoining equipment room. The 6-and 9-month procedures were identical and consisted of three parts. First, the Face-to-face procedure (adapted from Tronick et al., 1980) was performed. The procedure consisted of four two-minute episodes. The mother was seated in a chair, 2 feet away from her infant who was placed, facing her, in a car seat ( high-chair is used with larger infants.). The mother wa s instructed to play with her infant. She was free to interact with the child in any way, including physical touching. However, she was asked not to use toys, and not to take the child out of the seat. The two-minute free-play episode was followed by a still-face episode. The mother was instructed to sit back and remain unresponsive to the infant for the next two minutes. Next was a "reunion", or another free play period followed by a two minute play with "stranger". An unfamiliar adult replaced the mother in her seat and interacted with the infant for the remaining two minutes while the mother observed from the adjoining room. The procedure was abbreviated at any point the child became excessively upset and/or when the mother chose to performed. This 1s a structured 23-minute paradigm involving a series of separations and reunions of the mother and child, in addition to episodic interactions with an unfamiliar adult, "stranger".
Specified behaviors were coded from videotapes to yield a security of attachment classification for each child (Ainsworth, et al., 1978).
All scoring for the attachment classification were done by the Principal Investigator of the Infant Temperament Project.
Second, the children were presented with a series of abovedevelopmental-level toys to assess task-orientation.
As in the earlier visits, the final component of the 12-month lab visit consisted of 5 minute segments of each of the behaviors observed in the home.
The Attachment Q-sort: Each of the observers who visited the family weekly for the 8 to 10 months of the study, completed the attachment Q-sort following the child's first birthday. As a result of their extensive experience with the child and mother, these observers were quite familiar with the types of behaviors rated by this instrument. Each observer sorted the 90 behavioral statements into nine piles ( 10 statements each) according to how closely each statement represented the usual behavior of the child. The completed sorts were then compared to the criterion sort of the prototypically secure child to generate a security of attachment rating for each child.

Green
Scoring of maternal sensitivity was done by v1ewmg the first 10 minutes of the mother and child playing together. The remaining two situations (child playing alone and caretaking) were not scored.
because: (1) an initial survey of the data revealed that mothers were often not visible on tape during caretaking and play-alone episodes, and (2), the PCIS is best-suited for scoring interactive play between mother and child. This sconng was done by two reliable raters, who remained "blind" to the attachment classifications of the subjects.
Laboratory Observations of Maternal Behavior: The Face-to Face procedure conducted at the 6 and 9 month laboratory visit was used to compare the interaction style of each provided to ensure reliable coding. In this study, both episodes of mother-child free-play (total time-4 minutes) were scored. Each tape was scored twice, by two independent raters, who were unaware of the attachment classifications of the subjects.

Attachment
Procedures: The Strange Situation conducted at the 12-month lab visit was scored for security of attachment classification by an independent rater who had no knowledge of the maternal sensitivity scores.
Attachment classifications are derived by using detailed behavioral observations, particularly during the reunion episodes between mother and child (Ainsworth, 1978). Children who greet, make positive bids, smiles, or approaches towards their mothers are usually considered securely attached. Children who snub their mothers, by turning away, backing away or ignoring the mother's return generally fall into the avoidant (A) category. Children who express ambivalence by reaching toward mother, but then push away and otherwise resist physical contact, are classified as resistant (C).
Q-sorts of infant attachment were done by independent raters, "blind" to the purposes of this study. Each observer sorted the 90 behavioral statements into nine piles (10 statements each) according to how closely each statement represented the usual behavior of the child. The completed sorts were then compared to the criterion sort of the prototypically secure child to generate a security of attachment rating for each child.

Results of Reliability:
Rater reliability was established before final sconng began.
Three sets of 10 tapes were reviewed by three raters (the author and two undergraduate psychology students) and acceptable reliabilities on each of the 13 PCIS scales, as well as the total of the 5 Home Observations: Data reduction of the scoring of videotaped free play interaction yielded 3 summary variables for each mother-child dyad. These variables at 6 and 9 months were: Amount of Maternal involvement (AMNT6; AMNT9), Quality of Maternal Involvement (QUAL6; QUAL9) and Appropriateness of Maternal Involvement (APPR6; APPR9). Descriptive information for these variables (means and standard deviations) is provided in Table 2. Correlations among these scales at both 6 and 9 months are presented in Table 3.
Insert Tables 2 and 3 here As discussed above, there were 6 observation periods for each dyad around each age level (6 and 9 months This pattern of results is to be expected, because "amount" refers to the frequency of maternal response, whereas "appropriateness" and "quality" are both meant to assess the relative sensitivity with which responses are delivered. Due to high levels of consistency m these measures over time, final summary measures were derived by collapsing across the two ages to give 3 indices of the overall maternal style for each subject (AWIT, APPR, and QUAL).

Laboratory observations:
There were two Face-to -Face laboratory observations for each subject (6 and 9 months). Each session consisted of two 2-minute interactions between infant and mother. These 2 episodes were first scored separately and then collapsed across the two periods to generate the following measures: Sensitivity/Elaboration; Dyadic Regulation and Amount of Joint Activity at 6 and 9 months . In addition to these subscale ratings, global scores of maternal affect and overall quality of interaction were also made at 6 and 9 months . information for all these summary variables is provided in Table 5.
Insert Table 5 here In final analyses the following variables were used: SENSE6 ,

SENSE9, SENSE, as well as global measures of Maternal Affect and
Quality of Dyadic Interaction at 6 and 9 months. Correlations among these variables were calculated and are prese nted in Table 6. The summary sensitivity ratings as well as the global ratings were highly correlated across the 6 and 9 month assessments, with values ranging from r= .31 ; p<.10 to r=.94; p<.00.
Insert Table 6 here Maternal Behavior in Home and Laboratory Settin2s: To address the question of how multiple measures of maternal style in the home relate to one-time evaluations in the laboratory, laboratory measures were correlated with the two sets of summary variables from home observations. These analyses allowed for comparison of maternal style in naturalistic vs . structured environments and procedures at both age levels. These correlations are presented in Table 7.
Insert Table 7 here Generally, all ratings of quality and appropriateness m the home were significantly and positively related to all laboratory ratings of sensitivity. The rating of "amount of involvement" in the home at 9 months was significantly related to concurrent measures of sensitivity as well as concurrent global ratings . It was not significantly related to global ratings in the laboratory at 6 months.
The same rating of "amount of involvement" at 6 months was significantly related to laboratory measures of concurrent sensitivity, and global assessments of maternal affect at both ages. These home and laboratory correlations (primarily in the .40 to .50 range) indicated a moderate level of consistency between two sets of sensitivity measures.

Measures of Attachment:
Three outcome measures of attachment security were used.
For the purposes of analysis, the anxious-avoidant and the anxiousresistant groups were combined to yield a two -group classification variable (SECURE) . As mentioned earlier, higher incidence of secure classification in middle-class samples, often necessitates this method for approximating equal-n designs (Vaughn and Waters, 1990).
The study was originally designed to include an N=30 (selected from the larger sample N=49) with 10 subjects representing each of the attachment classification groups . However, of the 49 infants classified, 11 or 22% were "A", 29 or 59% were "B", and only 8 or 8% were classified as "C". This distribution is typical for a middle-class, normative sample. To approximate an equal-n design, it was therefore necessary to expand the sample size of this study . Of the 33 participants in this study, 10 were classified as"A"; 15 as "B" and Similarly, when summary variables derived from laboratory assessments were used for the same analyses, again no significant differences among attachment classification groups were found . These findings are presented in Table 8.
Insert However, when Amount was entered first and the set of Appropriateness and Quality was entered second, significant increases in R-squared were obtained . (R-squared change = .17, p< .05 at 6 months; R-squared change = .23, p< .01 at 9 months).
These findings are presented in Table 9.
Insert Table 9 here It is evident from this series of analyses that Appropriateness and Quality, variables assessing relative sensitivity of the maternal response, and not the frequency of responding, as assessed by Amount, contributed the most to predicting Q-sort attachment security .
Following similar logic , the question of whether laboratory assessments were related to attachment security measured by the Qsort, correlations between laboratory-derived summary measures of sensitivity and the Attachment Q-sort were computed. None of those comparisons proved to be significant or worthy of further investigation.
It should be noted that the results of this study will need to be interpreted cautiously given the small number of subjects currently available.
Findings are considered to be exploratory in nature.

Discussion
This study aimed to address the following major questions:

Can a reliable measure of maternal sensitivity be obtained?
The Parental/Caregiver Involvement Scale (PCIS) was used to assess maternal behavior in repeated, weekly free-play observations.
Inter-rater reliability of individual sessions was established to anticipated levels. Aggregated measures (over 6 weekly sessions) also proved to be highly reliable, even though week-to-week stability of any two sessions was modest. In addition, analyses of summary variables showed a high level of consistency within all measures, both within and across age level. These findings indicate that aggregation methods are extremely useful in obtaining a methodologically sound measure, which accurately reflects and incorporates the complexities of human behavior over time.
Laboratory Face-to-Face procedures were scored usmg a system adapted from Tronick (in progress). Inter-rater reliability was more difficult to establish with this system; two coders were required to obtain acceptable levels of reliability. Several explanations can be considered . First, the Face-to-Face procedure 1s a highly structured, fairly unnatural situation for both mother and infant. Infants are not normally placed in high-chairs to play, but to eat. Mothers typically engage in play with their infants by holding , cuddling, or moving freely on the floor and using toys for assistance.
None of these options are made available to mothers when they engage in the Face-to-Face procedure. Second, this laboratory paradigm allows for only 4 minutes of interaction between mother and infant which is used for sconng. It is quite possible that such a short time interval, in an unfamiliar , unusua l setting and structure makes it more difficult to apply the construc t of sensitive responsivity to maternal behavior. However, when two raters were aggregated, estimated reliabilities exceeded .80.

How are laboratory measures and home observations related?
Results of this study showed that measures collected m the home and in the laboratory were interrelated. Analyses demonstrated a positive relationship for most measures across age levels and across situations. This is an interesting finding given that separate measures were used to assess each situation. In other words, the PCIS, a system specifically designed for periods of longer, free-play interaction was appropriate for tapping maternal style under those circumstances. The Face-to-Face Scoring System was designed specifically with the constraints of the situation m mind.
Nevertheless, these distinct assessment tools were clearly useful in picking up aspects of the underlying construct of maternal style. When the Q-sort measure of attachment security was substituted for the Strange situation Classification, as an outcome measure of attachment, a different pattern of findings emerged.

4-~
Home-based measures of sensitivity were significantly related to the security of attachment, while laboratory measures were not.
Interestingly, both the Q-sort and the home-based ratings of sensitivity are measures collected over time, while laboratory assessments of attachment in the Strange Situation and of sensit ivity in the Face-to-Face paradigm take place during a single observation period.
It may be possible that information collected and aggregated over time more closely reflects the complexities of the constructs under consideration, or simply measures the constructs more reliably.
One-time observation in unnaturalistic settings may be affected by situational, confounding variables. Goldsmith & Alansky (1987), for example suggest that the Strange Situation is extremely sensitive to isolated incidents of resistant or avoidant behavior, and that it may not be applicable to children with higher fear/anger thresholds, as the situation would not be sufficiently stressful to trigger the attachment system. These authors conclude that the Strange Situation "may not provide the most sensitive measurement for reflec ting prior influences" . In a similar vein, Waters & Deane (1985) advocate Q-sort use for its ability to incorporate the context of behavior into its methodology, as well as minimize observer bias.
Generally, Q-sort measures reflect naturalistic behaviors, over time as judged by a knowledgeable observer. This approach parallels our assessment method of maternal sensitivity in the home.
Perhaps then, it is not surprising that these measures are more closely related than traditional attachment classification outcomes.

Does age level of the child at time of assessment
influence the relation of to maternal style and subsequent attachment classification?
According to Attachment theory, it is the accumulated expenence over the first year that results in a given attachment classification. It was expected that the 6 and 9 month data would explain a comparable portion of the variance m attachment classification, but one would not be more predictive than the other.
Results showed that both age levels were equally unrelated to attachment outcomes in the Strange Situation.
However, both the 6 and 9 month data were similarly related to attachment security as evaluated by the Q-sort. More specifically, measures of Quality and Appropriateness of maternal response, at both ages were correlated to attachment security. It appears that judgements made about a child's behavior by an observer, familiar with both the child and mother in naturalistic environment over time, are related to qualitative judgements of maternal behaviors, under similar conditions. In accordance with Attachment theory, one age level was not significantly superior to another in predicting attachment outcomes.

Comparison to Previous Findings:
Findings in this study were discrepant with some previous work linking maternal sensitivity and attachment classification outcomes. Studies reviewed have reported significant positive relationships between various measures of sensitivity during the course of the first year and attachment classification at twelve months (Ainsworth et al.,1978;Egeland & Farber, 1984;Belsky, Rovine and Taylor, 1984;Benn, 1985;Grossmann et al., 1985;Crockenberg and McCluskey, 1985;Smith and Pederson, 1988;Pederson et al., 1989;Isabella Belsky and VonEye, 1989;Isabella and Belsky, 1991;Morc:tn et al., in press). A summary review of these studies is presented in Table 1.

6.
Is methodology of this study inadequate in some way as compared to previous work?
In the design of this study careful consideration was given to issues of observer bias, reliability of measures , and assessment procedures. Attachment classifications were done by a trained and reliable, independent rater. Attachment classification distribution for the sample was comparable to previous work with middle class samples, and Ainsworth's original study (Ainsworth, 1978). As described above, measures of maternal style, both in the home and laboratory were carefully selected to reflect the constraints of those situations; they were also found to be highly reliable and consistent over time. All ratings were made by observers unaware of attachment classifications. Q-sort ratings were done by iqdependent observers, unfamiliar with the purposes of this study, based on their almost year-long acquaintance with the child and mother in question.
These methodological characteristics, then do not seem to be a probable explanation for our contradictory findings.
One clearly limiting factor of this study, is its sample size. As mentioned above, all findings need to be considered exploratory in nature, until replication with larger samples becomes possible. It should be mentioned, however, that several of the studies reviewed reported comparable sample sizes (Ainsworth, 1978;Benn, 1985;Pederson, 1989;Moran et al., in press).
Some strengths of the current study vis-a-vis previous studies are noteworthy.
Of the studies reviewed, few used repeated measures, and none, with the exception of Ainsworth's classic work, (Ainswort h, 1978) employed extensive, repeated home visits at various ages. The assessment measures used in this study , were also less subjective than those used by Ainsworth. In her original research, subjective observers, described as "semi-participants" rn the lives of the subjects constructed narrative accounts, on which all further rating was based . Although the objectivity of ratings has varied in recent research, the combination of theoretically and methodologically sound assessment measures, with repeated assessments in the home is unique to this study.
Finally, some researchers, (Goldsmith & Alansky, 1987) have pointed out that the often cited relationship between maternal sensitivity and attachment classification is not as robust as may be expected. Their metanalysis of studies revealed a weak, inconsistent effect for the studies reviewed.
Multiple assessments of attachment security were available for examination in this study. Although the Strange Situation and Q-sort measures were found to be related, (in this study, as well as in previous work [Waters & Deanne, 1985]) only the Q-sort had significant relationship to the measures of maternal style in this study. These findings are consistent with recent research (Pederson et al., 1989). In addition, the Attachment Q-sort has received attention in recent literature as a more dependable, context-based measure of attachment (Waters and Deanne, 1985;Goldsmith and Alansky, 1987;Smith & Pederson, 1988). The findings in this study support the notion of the Q-sort as a more sensitive, and methodologically sound measure of infant attachment, at least with regard to its relationship with maternal sensitivity.

Conclusions/Future Directions in Research:
These findings are by no means conclusive evidence against the link between maternal sensitivity and subsequent attachment security -an underlying assumption, central to current attachment theory. Rather, our findings point to the complexity qt_ the concepts _µnder study. The multi-faceted nature of dyadic interaction as well as individual behavior style , may call for intensive, prolonged observation. It may not be possible to make meaningful statements about these constructs after a single laboratoy-based procedure.
As noted above, this study will need to be replicated using l~ger ~~n:!QJ -2 izes to validate _ its findings. In addition, many important variables worthy of investigation were not examined in this study. Individual variables pertaining to the mother as well as ----------·---~the child need to be studied. For example, the contribution of the --- a11achmcn1 and maternal Q-sorts related SUMMARY: Acceptable reliabilities between raters reported, but each rater responsible for both measures for each subject; possibility of observer bias.
Factors associated with security of attachment in dual career families.
Society for Research in Child Development, Toronto, May, 1985.