Stability and Change in Infant Attachment in a Low_Income Sample

Research on attachment across infancy suggests, on the one hand, that there can be stability of attachment both within infancy (Owen, Easterbrooks, Chase_Lansdale, & Goldberg, 1984; Rauh, Ziegenhain, Muller, & Wijnroks, in press; Waters, 1978) and from infancy to early childhood (Howes & Hamilton, 1992; Main & Cassidy, 1988; Wartner, Grossmann, Fremmer_Bombik, & Suess, 1994). But it also suggests that instability of attachment, which appears to be more common than originally documented (Belsky, Campbell, Cohn, & Moore, 1996; Goldberg et al., under review), may be "lawful" and predictable (Egeland & Farber, 1984; Rauh et al., in press; Thompson, Lamb, & Estes, 1982; Vaughn, Egeland, Waters, & Sroufe, 1979). Change in attachment patterns is associated with changes in caregiving and with maternal and ecological risk factors that themselves can predict caregiving, though usually not in synchrony with changes in attachment (Spieker & Booth, 1988). Understanding not only what individual, family, and even community circumstances are associated with change in attachment, but also how they appear to operate is critical for understanding the ecology of parenting, parent_child relations, and child development.

The purpose of this investigation was to study the power of various individual and ecological factors (maternal characteristics and behavior, perceptions of social support, and infant characteristics) to differentiate early trajectories toward security (B), organized insecurity (A or C), and disorganization (D) among a sample of socioeconomically disadvantaged families. It was also designed to test the relative stability of attachment in a low_income sample and the relation between these various ecological factors and the level of disorganization or atypicality observed in each attachment. Thus, there is a focus on both traditional patterns of attachment as well as level of disorganization in attachment behavior, and the relation of each to a variety of maternal, child, and contextual factors.

Maternal behavior and maternal and ecological risk factors have already been found to explain higher rates of insecure infant_mother attachment at one point in time. Theoretically important categories of variables that Belsky (1984; Vondra & Belsky, 1993) proposed should influence the quality of parenting have, in fact, been associated with infant attachment security. The vast majority of these studies have been conducted with white, middle_class families, and have used the traditional A,B,C attachment classification system (Ainsworth, Blehar, Waters, & Wall, 1978). However, research on high risk or dysfunctional parents has replicated and extended these findings. Summarizing results from seven programs of research involving attachment studies of high_risk, mentally ill, and/or maltreating families, Spieker and Booth (1988) concluded that adolescent parenting, low educational attainment, perceptions of low social support, maternal defensiveness and/or depressive symptomatology, and "insensitive" maternal behavior (whether hostile, unavailable, or intrusive/controlling) all are associated with insecure attachment. To some extent, these are also factors that explain changes toward insecurity over time, usually assessed between infancy (12_14 months) and toddlerhood (18_22 months).

Although much of the research supporting hypotheses about stability (and occasionally change) in attachment security has been based on samples of low_risk, middle_class, mother_child dyads, a handful of studies have investigated stability of attachment across infancy among low_income and/or high_risk populations. These studies suggest, first, that changes in attachment security__particularly increases in disorganized or atypical ("Unstable A" or "A/C" or "Unclassifiable") attachments__are more characteristic of disadvantaged, high_risk, and/or maltreating populations than they are of low_risk, middle_class populations (Egeland & Sroufe, 1981b; Lyons_Ruth, Repacholi, McLeod, & Silva, 1991; Spieker & Booth, 1985; Vaughn et al., 1979). They also indicate that among the most consistent correlates of the emergence of insecure patterns by 18 months are maternal negative affectivity and psychologically unavailable or hostile/controlling behavior.

Studies involving only classification into the traditional (organized) A, B, and C categories reveal that negative affect or unavailability on the part of mothers is associated with stable insecurity or movement toward insecurity by the second year of life. In most cases, the predominant insecure pattern is avoidance (resistant attachments occur with less frequency, and usually cannot be analyzed separately). Negative affect has been assessed in terms of depressive symptomatology (Egeland & Sroufe, 1981a, 1981b) and a combination of self_reported aggressiveness, suspiciousness, and low social desirability (Egeland & Farber, 1984). Both maternal unavailability (Egeland & Erickson, 1987) and punitive attitudes or controlling behavior (Frodi, Grolnick, & Bridges, 1985) have been studied. Each of these indices of maternal functioning was assessed in the present investigation. Using the D classification, both Egeland and Sroufe (1981b) and Lyons_Ruth and her colleagues (1991) found that psychologically dysfunctional mothers (i.e., those with psychiatric hospitalizations or evidence of child maltreatment) had infants who moved to an atypical classification (usually force_classified as insecure) by 18 months. Barnett, Ganiban, and Cicchetti (this volume) report a high percentage of D classifications beginning at 12 months among a sample of maltreated toddlers and relatively greater stability of classification, perhaps due to the high percentage of type D attachments at every age. A small number of families in the present study were reported for child maltreatment, either of the index child (n=6) or a sibling (n=2), permitting some anecdotal information regarding maltreatment and disorganized attachment.

In terms of infant characteristics, studies show higher physiological reactivity among infants with D attachments in the Strange Situation (Hertsgaard, Gunnar, Erickson, & Machmias, 1995; Spangler & Grossmann, 1993), although their observed distress isn't significantly different from that of infants with other attachments (Barnett et al., this volume). Prospective studies on observed or mother_rated temperament of infants with atypical attachments have not been done. The extant literature suggests that these infants are not necessarily more negative in affect during early infancy, but become easily aroused and/or less able to return to baseline arousal__at least in a separation context__over the course of their early experiences. Their observed distress in the Strange Situation subsequently increases over the second year (Barnett et al., this volume). Whether this increase in distress occurs outside a separation context is unknown. Both observer ratings of fussiness in the lab and maternal ratings of temperamental difficultness were gathered on repeat occasions in the present investigation, allowing examination of these questions.

Given the greater frequency of insecure and disorganized or atyical attachment patterns in at_risk samples, longitudinal studies of attachment in children from disadvantaged and/or troubled families have two distinct, but related, issues to address. First, are there unique correlates of attachment trajectories that move toward versus away from security? Second, are there unique correlates of trajectories that move toward versus away from disorganized attachment patterns? In most cases, sample sizes have not been sufficient to examine these two issues simultaneously. Specifically, there are almost no data available adequate for analyses to distinguish infant_mother relationships that show increased avoidance, resistance, or disorganization, versus security. Since many infants from at_risk samples who are classified as atypical or disorganized show modest to moderate avoidance (and are force classified avoidant), it would be particularly useful to distinguish between infants showing movement over time toward organized avoidance versus disorganization. This would begin to build a bridge between older research using the traditional A,B,C scheme versus more recent work incorporating a disorganized or atypical category. An important question that has not yet been addressed empirically is whether a different set of ecological correlates distinguish attachments according to the A,B,C scheme versus attachments varying along a dimension of atypicality. The current study provides an opportunity to address this need, given the availability of longitudinal data on both organized and disorganized attachments for a sample of 90 mother_infant dyads from low_income, socioeconomically disadvantaged families.



Mother_infant dyads for this study were drawn from a sample of 103 urban, low_income mothers and their one_year_old infants participating in a longitudinal investigation of vulnerability and resilience in early childhood. Mothers with infants between the ages of 5 and 11 months were recruited for the larger investigation from the Women, Infant, and Children (WIC) Supplemental Nutrition Program administered by the County Health Department. Inclusion in the WIC Program requires low income (e.g., less than $22,385 for a family of four, less than $14,837 for a single mother and child in 1989, the first year of study recruitment). Women contacted through the WIC office who completed brief background questionnaires and took part in an initial lab assessment with their infant at 12 months were included in the longitudinal investigation. Seventy_two percent of the women approached at WIC came to the University for the first research visit. Only 5% of those contacted at WIC refused to participate and an additional 23% never appeared. Because screening data were collected at WIC on mothers in the first cohort who agreed to participate, it was possible to test for differences between mothers who did or did not appear for their first visit, or completed only two of four infant assessments. No differences emerged on maternal age, education, race, marital status, or number of children, on self_reported personality, or on perceptions of infant difficultness. Of the 103 mother_infant dyads who appeared for a lab visit, 90 had attachment data at both the initial 12_month visit and a follow_up, 18_month visit. These 90 dyads comprise the sample for the present investigation.

At the first lab visit (12 months), 41% of mothers reported being married or living with a partner, 14% reported being separated or divorced, and 45% reported being single. Three quarters of the women (75%) reported having a high school education or less at the time of recruitment (19% had no high school degree), 89% reported having a family income of less than $1,500 per month, 40% were of minority race (1 Hispanic, 35 African_American), and 17% were teenagers at the time of their first child's birth. Mothers were asked about child maltreatment occuring in their home and six mothers either reported that the target child had been a victim of maltreatment or subsequently had their child removed from the home by Child Protective Services. Two additional mothers reported maltreatment of the child's sibling. Maternal age at the time of recruitment ranged from 16 to 37, with a mean of 25. Of the 90 infants 40% (36) are female, 47% (42) are firstborn, and 12% (11) were born more than two weeks prematurely. The date of the assessments was corrected for infant gestational age.


Mothers were paid for completion of the recruitment instruments at the WIC office and for completion of two lab visits at the University and one home visit (at 15 months). Each lab assessment (12 and 18 months) began with a free play period for the infant (when mothers worked on questionnaires with the Examiner), followed by a series of interactive activities (Examiner absent from the room), a rest/snack period in another room, and the Strange Situation. Mothers then completed questionnaires with the Examiner while a staffmember entertained the infant. The lab visits took approximately two hours to complete and were videotaped through a one_way mirror. The home visit was conducted when infants were approximately 15 months old, and consisted of an infant developmental assessment, an infant free play period (when mothers completed questionnaires with the Examiner), and an interview/observation period during which the quality of the home environment as a context for child development was assessed.


The purpose of the present investigation was to examine attachment stability and change__and factors associated with each__from 12 to 18 months of age. A primary goal was to distinguish infants moving toward versus away from attachment security (i.e., on trajectories toward socioemotional competence or vulnerability) and moving toward versus away from organized attachment strategies (i.e., showing any disorganization across infancy). Measure selection was guided, in part, by the work of Egeland and his colleagues (Egeland & Farber, 1984; Brunnquell, Crichton, & Egeland, 1981) on attachment among a large sample of infants from disadvantaged backgrounds, and attachment coding strategies were guided by the growing body of work on attachment disorganization and atypicality among infants at risk for developmental psychopathology.

The measures used for this study are summarized in Tables 1 and 2. In some instances, data were combined across two or three assessments to capture a broader index of individual or family circumstances over time. Whenever possible, repeated measures were also examined in terms of change scores from 12 to 18 months.

Infant_Mother Attachment Classification. Attachment security was assessed using the Strange Situation paradigm (Ainsworth & Wittig, 1969). Three coders were trained to reliability (within this lab) and were also separately tested for interrater agreement (outside this lab) using two different sets of attachment assessments, one set from the lab of J. Belsky and a second set from the lab of A. Sroufe. The principal coder, J. Vondra, was trained in the A,B,C system (Ainsworth, Blehar, Waters, & Wall, 1978) by Mary Jo Ward and Brian Vaughn, and in the D system (Main & Solomon, 1990) by Dante Cicchetti; the two additional coders were trained by Vondra.. The principal coder scored 84 (all but six) attachments at 12 months and 30 attachments at 18 months (generally attachments where there was disagreement between coders). All other tapes were scored by one or both additional coders. Of the 28 tapes given a final rating by the principal coder at both 12 and 18 months (with scoring always separated by one to two years), stability of classification was somewhat less (36%) than overall study stability (50%), with stability similarly concentrated among Type B and Type D attachments. Mean interrater reliability (in this lab) on major classifications was 80% for all combinations of the three raters using a random set of 10 assessments (representing all four major classifications) from the current study. External agreement on A, B, and C classifications ranged from 80% to 100%, with a mean of 86%, across the three raters for the test assessments.

Because D training took place before publication of the D system, when the D_scale had not yet been developed (a seven_step decision process was outlined), coding of disorganized attachment was conducted conservatively in this investigation, as described below. Raters looked for two sets of features: (1) behavioral indices of disorganization or depressed affect as described by Main and Solomon (1990)__for example, covert swipes at mother, backward or sideways approaches to mother, backing up when mother approaches, simultaneously moving toward and away from mother__and (2) disorganized behavioral sequences across the Strange Situation episodes__for example, marked distress and search for mother in the stranger's presence, followed by immediate avoidance of mother at reunion; intense avoidance in the first reunion with mild or marked resistance but no avoidance in the second reunion. An attachment was classified as disorganized only when (1) there was an overwhelming number of behavioral indices of disorganization or depressed affect in the absence of disorganized behavioral sequences (i.e., an organized classification could be identified, but multiple behavioral indices of disorganization were observed simultaneously) or (2) there were at least some behavioral indices of disorganization but there was also disorganization of behavioral sequences so that a traditional classification could not be readily identified (A/C), or each reunion had a distinct__and opposing__classification (usually A then B4, or A then C, with a drop of 4 or more points in avoidance, the Unstable A designation). In contrast to the Main and Solomon (1990) criteria, no single behaviors were considered sufficient__in and of themselves__for inclusion in the disorganized category. However, the presence of a small number of indices of disorganization in the absence of disorganization in the behavioral sequences was noted by use of the designation, "D_signs." This would be roughly comparable to a 3 to 5 on the 9_point D_scale (Main & Solomon, 1990). The D classification would be roughly comparable to a score of 6 to 9 on the D_scale. Infants were designated as stable in having a disorganized attachment when they were classified as disorganized in at least one assessment, but also showed some behavioral indices of disorganization at the second assessment, whether or not they were formally classified as "disorganized." Of the 12 cases designated as stable D, 5 were classified D at both assessments, 4 were coded as having D_signs at both assessments, and 3 showed D_signs at 12 months and were classified D at 18 months.

Demographic Risk. A summary risk score for each mother was computed on the basis of demographic information collected during the initial recruitment session. Contributing factors (scored as zero or one, and summed) were teen parenthood at first birth, failure to complete high school, and racial minority status. Each of these indices of demographic status is associated with socioeconomic hardship and lower social status in the United States, important risk factors for individual and family functioning.

Maternal I.Q.. Maternal IQ was estimated from two subscales of the WAIS_R, administered when infants were 24 months old. The composite score from these two subscales, Vocabulary and Block Design, has a high correlation with the full_scale IQ (Sattler, 1988).

Personality Risk. Based on Egeland and Farber's (1984) work, a personality risk score was created from subscales of a personality inventory (the Personality Research Form, Jackson, 1989) administered both at recruitment (5_11 months) and at the 18_month lab visit. The three scales each include 16 true_false items, with internal consistency ranging from .72 to .87 (Stumpf, Wieck, & Jackson, 1976), and test_retest reliability ranging from .84 to .87 (Jackson & Morf, 1973). Mothers scoring above the sample mean on both aggressiveness and defendence, and below the sample mean on social desirability, were considered at risk of fostering an insecure attachment relationship with their infant. Risk scores (zero or one) were summed across the two assessments in an effort to create a more reliable measure by capturing more enduring personality dispositions. Final scores therefore ranged from 0 (never fit the personality profile) to 2 (fit the personality profile at both assessments).

Anger Control. Maternal self_reported anger control versus anger expression was obtained during the home visit at 15 months. This variable was computed as the difference between anger control and anger expression scores on the Anger Expression Scale of the State_Trait Anger Expression Inventory (Spielberger, 1991). Mothers scoring higher on this measure endorse items indicating ability to control anger ("I keep my cool"), but not items indicating behavioral expressions of anger ("I argue with others").

Depressive Symptomatology. Self_reported maternal depressive symptomatology was assessed at 12 and 18 months using the short form of the Beck Depression Inventory (Beck & Beamesderfer, 1974). Scores were both subtracted (12 _ 18 months), to create a measure of change in depressive symptoms, and summed (12 + 18 months), to capture average level of depressive symptomatology over the six_month study period.

Relationship Satisfaction. Maternal perceptions of the quality of her relationship with a partner were assessed using the short form of the Marital Adjustment Test (Locke & Wallace, 1959). This measure has proven successful in discriminating harmonious and disturbed/conflicted marriages (Gottman, Markman, & Notarius, 1977; Hershorn & Rosenbaum, 1983; Rosenbaum & O'Leary, 1981), and also predicts children's behavior problems (Emery & O'Leary, 1982). Scores on the measure correlate with self_reported relationship violence and with children's distress reaction to expressed adult anger (Cummings, Pellegrini, Notarius, & Cummings, 1989). Because a number of mothers reported that a person other than a husband or boyfriend served as their significant other (16% mother, 9% other relative or friend, 11% no significant other at 12 months), the measure's wording was modified ("relationship" versus "marriage") for use with whatever partner (mother, sister, girlfriend) might be identified. Scores at the 12_ and 18_month lab visits were once again combined in two ways: (1) difference scores were computed to measure change in level of satisfaction, and (2) sums were computed to capture the average quality of partner relationship over the period of study.

Disruptive Events and Influences. Other contextual influences of importance were examined through a summary score of events and influences that mothers reported during the period from 12 to 18 months. This score was the sum of five measures, each scored present or absent: (1) negative life events reported at the 15_month visit above the sample mean (using the Life Experiences Survey, Sarason, Johnson, & Siegel, 1978), (2) more than two changes in childcare arrangements reported between 12 and 18 months (e.g, from maternal care to grandmother care to babysitter), (3) mothers' report of only negative family influences on her child or her denial of anyfamily influence during this same period (it was assumed that denial of family influence is an indication that mothers will fail to see a need to protect their child from problematic family events or circumstances), (4) the birth of a sibling before the target child turned 18 months, and (5) two or more moves in the child's second year of life. Total scores reflect disruptive, stressful events occurring before the children turned two.

Infant Difficultness. Two kinds of assessments for infant difficultness were obtained to distinguish between observed infant behavior and maternal perceptions of difficultness. Behavioral measures consisted of (1) the sum at 12 and 18 months of the number of seconds infants fussed during the 15_minute free play, and at 12 months, the number of seconds infants demanded maternal attention during a brief period sitting in a high chair (most typically by fussing) and (2) thedifference in amount (seconds) of fussing at the 12_ versus 18_month free play. The perceptual measures were (1) the sum of ratings of difficultness mothers made at the screening, the 12_month, and 18_month visits, using items constituting the "difficultness" factor for each age on the Infant Characteristics Questionnaire (Bates, Freeland, & Lounsbury, 1979), and (2) the change in ratings from the screening to the 12_ and 18_month visits (S _ 12 _ 18 difference score). Interrater reliability for infant demandingness was calculated on 14 cases selected at random from the sample as the mean percent agreement among three coders who were blind to attachment classification of incidences of each behavioral code for three_second scoring intervals. Mean agreement was 93%. Interrater reliability for fussiness during play was calculated on 60 cases at 12 months and 20 cases at 18 months. Percent agreement between raters about the occurrence of any fussing across the entire play session was 85% and 75%, respectively. Among children with any coded fussing, agreement about the amount of fussing (in seconds) was computed by correlation. Correlations were .98 at both time points.

Small correlations (around .20, p < .10) were found between the three observed indices of fussing/demanding, moderate to strong correlations (.51 to .70, p < .001) correlations were found between the three maternal reports of difficultness, but no correlations were found between observed fussing and reported difficultness. The authors of the ICQ reported moderate correlations between mothers' and observers' ratings on the ICQ (r[98]=.34, p < .001) and acceptable test_retest stability (r=.59) between observations (2_10 day intervals).

Quality of Maternal Care. Measures of maternal care included both behavioral ratings (Ainsworth et al., 1978) and behavioral frequencies (Martin, 1981; Smith & Pederson, 1988) assigned to mothers by trained raters blind to attachment classification during three different interactive tasks during the lab visit at 12 months. In the first, infants were seated in a high chair for three minutes while mothers worked on questionnaires nearby; in the second, infants were removed from the high chair but given no toys to play with for an additional three minutes while mothers continued completing questionnaires; and in the third, mothers were asked to get their child to complete a series of challenging tasks (e.g., putting away toys, working an activity box, completing a puzzle) for a total of 12 minutes. Independent raters coded each of the three tasks. These caregiving tasks and scores are described in more detail elsewhere (Vondra, Shaw, & Kevenides, 1995), demonstrated adequate interrater reliability and, when used to create composite indices, discriminated among mothers of infants with avoidant, resistant, and secure attachment classifications at 12 months.

Quality of the Home Environment. Ratings of the quality of the home as a supportive context for child development were completed by observers at the 15_month home visit using theHome Observation for Measurement of the Environment (HOME) Inventory (Caldwell & Bradley, 1984). The infant version consists of a 45_item checklist arranged in six subscales, including "Emotional and Verbal Responsivity of Parent," "Acceptance of Child Behavior," and "Provision of Appropriate Play Materials." Information needed to score Inventory items is obtained through a combination of observation and interview (with the mother). The total score was used in analyses.

Results and Discussion

Attachment Stability

Table 3 provides data on stability and change in attachment classification from 12 to 18 months. At 12 months, 50% of the sample was scored secure, 30% organized insecure, and 20% disorganized. By 18 months, only 42% was scored secure, 29% organized insecure, and 29% disorganized. The number of secure attachments decreased and number of disorganized attachments increased across infancy. Stability in this sample, like that in Barnett's (this volume) maltreated sample, was concentrated among children with either secure or disorganized attachments. Twenty_nine percent of the 90 children were coded as having secure attachments at both 12 and 18 months; 13% were coded as showing disorganization in their attachment at both 12 and 18 months. Only 8% of the sample showed a stable pattern of organized insecurity (either avoidance or resistance). This represents a 50% rate of classification stability, a very modest but__with a sample of 90__statistically significant pattern (K = .28, p < .05). There was as much change in classification as there was stability over a six_month interval.

Table 4 provides a comparison of these results with four other investigations of infant_mother attachment among disadvantaged populations. Egeland and Farber (1984) reported only stability data within the A,B,C classification scheme. Both Lyons_Ruth et al. (1991) and Barnett, Ganniban, and Cicchetti (this volume) had smaller samples, about half of which consist of cases of child maltreatment. Maslin_Cole and Spieker (1990) did not provide overall stability figures. In terms of forced classifications, the four relevant investigations are quite comparable, with stability of classification ranging from 54 to 60%. However, there is great variability in terms of stability when disorganization is identified and distinguished from organized classifications (30_67%). The only figure in any way comparable across all three investigations, in that case, is the percentage of disorganization noted at 18 months (29_46%), which is higher than that van IJzendoorn, Schuengel, & Bakersman_Kranenburg (in press) reported (24%) in their metanalysis of low_SES samples. Although rates of disorganization were lowest in the current investigation (no doubt reflecting both the maltreatment sampling in other studies and the conservative scoring strategy adopted in the present study), they were comparable to two of the three other investigations. These data suggest sample variability, certainly, but also probable variation in scoring criteria for disorganization. Although there is reasonable agreement, based on the work of Main and Solomon, about the various behavioral indices indicative of disorganization, there is potential for disagreement about the number and severity of indices that must exist before a child's attachment is deemed "disorganized." This is captured in the broadly delineated "D scale" by Main and Solomon (1990), and by the distinction between no D signs, some D signs, or the D classification in the present study.

Attachment Disorganization

Among the 24 children whose attachment showed at least some signs of disorganization at 12 months, 13 were designated as having D_signs, 5 showed the A/C mix, 5 showed unstable avoidance, and 1 appeared depressed. Of these subsets of children, the A/C group was least likely to be force_classified as secure or to become secure at 18 months. Among the 35 children whose attachment showed at least some signs of disorganization at 18 months, 14 were designated as having D_signs, 7 showed the A/C mix, 6 showed unstable avoidance, and 8 appeared depressed. Of these subsets of children, the A/C group was again the least likely to be force_classified secure or to have been secureat 12 months. Disorganized attachments designated as A/C were somewhat less likely to be associated with security and more likely to be associated with disorganization than were those designated as Unstable A, probably due to the coding requirements of each (e.g., the Unstable A pattern requires a marked decrease in avoidance). Attachment history and force_classification of attachments showing some disorganization is summarized in Table 5.

Group Differences

In this section, data are summarized with respect to attachment group differences on each of the indices summarized in Table 2. Six attachment subgroups were identified for study that reflect two methodological objectives: (1) maintaining distinctions between different patterns of attachment across time and (2) ensuring adequate subsample size and internal consistency within groups. The six attachment subgroups designated in Table 6 seemed to provide the best balance between these two priorities. However, it should be noted that the relatively large number of subgroups (6) and the small group sizes (6 to 26 dyads) limit the power of the statistical tests. These consisted of univariate ANOVAs for each independent variable and Bonferoni tests for post_hoc comparisons of differences that emerged.

Given the relative stability of the secure and disorganized groups from 12 to 18 months, we distinguished between those infants who were consistent over time in their security ("Stable B") or their disorganization ("Stable D") and those who moved toward security ("A,C,D___> B") or disorganization (A,B,C___> D"). On the other hand, the relative instability of avoidant and resistant strategies forced us to combine infants into groups based on their movement toward organized avoidance ("A,B,C,D___> A") or organized resistance ("A,B,C,D___> C") in their attachment. As can be seen from the group difference results, summarized in Tables 7 and 8, this scheme proved fairly successful in providing some insights into individual, relational, and circumstantial differences between the mother_infant dyads we studied. Overall effects are summarized here, and differences between attachment subgroups are described subsequently.

Maternal Variables. Of the maternal variables, demographic and personality risk, self_reported anger control, and reported disruptive events and influences each showed an overall (sample_wide) effect of attachment subgroup. Summed maternal relationship satisfaction showed a trend toward significance due to differences related to one of the smaller subgroups. Only maternal depressive symptomatology showed no relation to subgroup membership. It is not the case, however, that depressive symptomatology was unrelated to changes in attachment. Separate analyses (Hommerding, Shaw, & Vondra, 1993) showed that mothers reporting greater symptomatology had infants who were more likely to move from a secure to any insecure pattern by 18 months, particularly girls. Self_reported depressive symptoms did not, however, distinguish among the trajectories toward the specific patterns of insecurity examined in the present study.

Finally, change scores based on both 12_ and 18_month data, computed for partner satisfaction and depressive symptomatology, did not distinguish among the attachment subgroups. The failure of the change scores to explain changes in attachment security may reflect methodological issues more than substantive ones. Summing 12_ and 18_month scores may improve the reliability of the measures significantly over change scores. Alternatively, changes in mothers' relationship perceptions and self_ratings may be relatively insignificant in the face of broad individual differences between mothers in these perceptions, and/or may not be apparent in terms of mother_infant relationship quality until some later point in time. It is worth noting, however, that five of the six (individual or summed) maternal variables under investigation showed some effect of infant_mother attachment trajectory in this analysis.

Infant Variables. Group effects were also found using the infant temperament variables, particularly observed behavior. Decreases in high observed infant fussiness from the first to the second year (12 months _ 18 months) helped distinguish the "C" attachment subgroup. Greater overall (summed) observed fussiness also showed this group difference, but mothers' summed rating of infant difficultness (screening + 12 months + 18 months) did not. Decreased difficultness of the group of children who became "C" by 18 months resulted from extremely high difficultness scores at 12 months, due to a subset of unusually fussy infants, and scores that were still high, but more consistent with the rest of the sample, by 18 months. Likewise, mothers in the "C" group reported the highest infant difficultness at screening (when infants were 6 to 12 months old), but lower difficultness at 12 and 18 months. Less sensitive care and insecure attachment have both been associated with greater behavioral and emotional "difficultness" across infancy (Engfer, 1986; Fish, Stifter, & Belsky, 1991; Matheny, 1986). Whether a baby with "difficult" temperamental characteristics elicits less sensitive care, or a parent who sees his or her infant in a negative light provides less sensitive care, there is certainly a basis for expecting some association between infant fussy, demanding behavior and both parental caregiving and infant_parent attachment security. At the same time, fussy, demanding infants may be somewhat more prone to developing anxious_resistant (Calkins & Fox, 1992; Crockenberg, 1981; Grossmann et al., 1985), or at least insecure (Type A, van den Boom, 1989), attachments, as some data suggest. In this sample, assessed difficultness among the children whose attachments moved to C at 18 months seemed more exclusively an issue in the first year of life.

Care Variables. Only the indices of care were consistent in showing no effect of attachment trajectory. Across all groups, measures of maternal care and the quality of the home environment, surprisingly, did not show any pattern of systematic covariance with attachment classification over time. Although 12_month measures were associated with contemporaneous, 12_month forced classifications (Vondra et al., 1995), there was no predictive association with attachment at 18 months using 12_month indices of sensitive care, negative behavior, or unresponsive behavior. In most cases, the variability within groups was quite large, easily overpowering variability between groups. The only consistent observable pattern was a tendency for the small number of mothers of infants who were classified as resistant by 18 months to look consistently best on 12_month care measures and the 15_month measure of home environment, similar to a finding reported by Egeland and Farber (1984). It should be noted, however, that none of these measures attempts to capture covert hostility (also termed affective communication errors), a contradictory blend of positive affect and hostile behavior, disorientation, or role reversal, which Lyons_Ruth and her colleagues (1991, this volume) found useful in distinguishing infants with disorganized attachments. On the other hand, both withdrawal and negative, intrusive behavior were assessed, and proved unsuccessful in differentiating attachment change groups. However, in the six cases of known child maltreatment involving the target child, all but one involved a "D" classification at 12 or 18 months, and the single exception was designated as stable D because D_signs were noted at both 12 and 18 months. One of the two sibling abuse cases was classified as D at 12 months (both moved to "A" at 18 months).

In the next section, these overall effects are examined in greater detail in terms of their ability to discriminate among the attachment subgroups. Rather than present the findings in piecemeal fashion, variable by variable, the findings are summarized for each subgroup. The point in doing this is to create a more integrated picture of the nature of and differences between mother_infant dyads comprising the six attachment trajectory groups. The relevant data appear in Tables 6_8.

Attachment Subgroups

Stable Security. Almost 60% of the infants who were secure in their attachment relationship at 12 months were also secure at 18 months. This group of mothers and infants was most remarkable for never scoring high on any risk factors. In this sample, the stable secure dyads stand out for what they lack by way of risk. In fact, this group was usually among the lowest on all mean risk indices, although there was enough subgroup variation across the whole sample that this difference was only occasionally significant. Relatively few mothers reported aggressive and suspicious feelings without a need to look socially desirable (significant), clinically significant depressive symptoms, either notably angry feelings or high anger control, or dissatisfaction with intimate relationships. They also reported fewer disruptive family events or influences (significant).

Trajectory toward Security. About one quarter of the infants insecurely attached at 12 months appeared secure by 18 months. Avoidant, resistant, and disorganized attachments at 12 months were equally represented (n=4 of each) in this group. In 75% of the cases, this was a first child, and there was some indication that these were more likely to be teen mothers, high school dropouts, and/or of minority race. They were also the least likely to have a spouse or cohabiting partner at both 12 and 18 months (only one third had one). But these mothers scored relatively lowon the aggressive/suspicious personality risk complex that Egeland and Farber (1984) identified, high on mean relationship satisfaction (with partner, friend, or mother), and low on disruptive events and influences on their infants in the first year. These may be women who became mothers before they were psychologically prepared for the role, but who were able__in the absence of too many personal or circumstantial stressors__to grow with the role across infancy.

Trajectory toward Organized Resistance. This was the smallest subgroup of the study, only 7% (n=6, two of whom showed stable resistance) of the sample, so differences associated with this group are most tenuous. Still, they provided the clearest replication of results Egeland and Farber (1984) reported for their low_income sample; namely, what appears to be a group of hard_to_care_for infants with mothers who grow increasingly anxious and depressed across infancy.

In the current sample, these infants were more often firstborn and male, with the highest sample mean on total observed difficultness, but also the greatest decline in fussing from 12 to 18 months. This was the only subgroup scoring high across all three measures of observed temperamental difficultness (two at 12 months, one at 18 months). Mothers did not, however, report that their infants were noticeably difficult, except at the initial screening when infants were between 6 and 12 months old. None of these women were at demographic risk due to age, education, or minority status, but instead were all white and had intimate, cohabiting partners both at 12 and 18 months. However, they were noticeably negative about their marital relationship. This group may represent the problematic combination of a struggling marriage and a baby who is, at least initially, more difficult in temperament.

Trajectory toward Organized Avoidance. By 18 months, 19% (n=20, 5 of whom showed stable avoidance) of the infants were classified as avoidant. These children were more often laterborn. Mothers in this group were least likely to have someone they could call a partner or significant other at 12 months, and half had no husband or cohabiting boyfriend at both points in time, a finding that echoes the pattern of not living with a partner discerned by Egeland and Farber (1984) in their "Stable Avoidant" group. This may be an important reflection of discontinuity of relationships among women in this group, who not only had had a partner a year or two earlier, but also tended to have older children as well.

The only other distinguishing characteristic noted for women in this group is their self_reported anger. On average, these mothers reported the most anger expression and the least anger control in our sample. However, both at 12 and at 18 months, these mothers were among the least likely to show anger toward their infants during the teaching tasks. The negative affectivity that both Belsky (Belsky & Nezworski, 1988) and Spieker (Spieker & Booth, 1988) have discussed in relation to attachment insecurity was expressed quite differently in the various subgroups of mothers of insecurely attached infants. Whereas the mothers of infants who moved toward resistance expressed feelings of sadness and of dissatisfaction with their close relationships, the mothers of infants who moved toward avoidance reported feelings of anger. A third pattern of negative affectivity emerged for the mothers of infants demonstrating a disorganized attachment, relating to personality factors and disruptive family events and influences.

Stable Disorganization and Trajectory toward Disorganization. Twenty_nine percent of the sample was classified as disorganized in attachment by 18 months of age, with half of that group showing either disorganized signs or patterns at 12 months as well. As Lyons_Ruth and her colleagues found (this volume) there may be an over_representation of boys in these subgroups (67_71% male, versus 40% male for the sample as a whole). Mothers scored highest on the personality risk complex that Egeland and Farber (1984) identified: high on aggressiveness, high on defendence or suspiciousness, and low on social desirability. A question worth raising is how many of the children in the "B"_to_"A" group in Egeland's sample, whose mothers scored high on this complex, would actually have been classified as disorganized at one or both assessments, using current standards.

Mothers of the 15% of children who moved out of an organized attachment strategy at 12 months to a disorganized attachment classification at 18 months reported the most disruptive family events and influences across the first year. These same mothers, however, reported having the leastexpressed anger and the most anger control at the home visit, although they showed no less anger toward their infants during teaching tasks at 12 and 18 months than did other mothers. At 12 months, their infants' attachments had largely reflected the sample distribution (50% "B," 21% "C," and 29% "A"). They described negative events and changes, talked about entirely negative or no family influences whatsoever on their infant, but also described themselves as people who do not get angry.

In sum, both groups of mothers whose infants were classified as disorganized in their attachment by 18 months seemed different on several personality indices. It is not, however, simply a case of their being higher on overall negative affectivity. Mothers whose infants began showing signs of disorganization only at 18 months tended to report many disruptive events or circumstances, but not the affective experiences one might expect to accompany them, neither depression, anger, nor dissatisfaction with their intimate relationship. Mothers whose infants were consistently disorganized were more likely to describe themselves and others as hostile, yet were satisfied in their relationship with a significant other (whether husband, friend, or mother). It is possible that this is another reflection of the kind of affective inconsistency Lyons_Ruth and her colleagues (1991, this volume) have captured behaviorally in mothers of infants with D attachments.

Analyses Within the Disorganized Group

Two kinds of subgroup analyses were run using only children with disorganized attachments by 18 months, but no consistent differences were found on any of the variables from Table 2. These analyses distinguished between (1) infants who were force classified as secure versus insecure and, separately, (2) infants who demonstrated a pattern of mixed avoidance and resistance in each reunion (A/C) versus high avoidance in the first reunion and no avoidance (but often resistance) in the second reunion (Unstable A). Means were so similar in both comparisons that sample size restrictions (and consequent lack of statistical power) did not appear to be an explanation. Thus, the distinction Lyons_Ruth and her colleagues (1991, this volume) found useful between infants with signs of disorganization who fit a "secure" versus insecure pattern did not prove effective here in distinguishing mothers on a variety of demographic and self_report measures. Similarly, mothers of children whose attachments combined avoidance and resistance in each reunion (A/C) or changed dramatically across reunions from avoidance to proximity_seeking with or without resistance (Unstable A) did not differ in describing themselves, their children, or their circumstances.

As noted earlier, however, the A/C combination was somewhat less likely to be associated with security and more likely to be associated with disorganization within and over time. Of the six child maltreatment cases, all of which showed some disorganization, one was designated as D_signs at both 12 and 18 months, one each showed the A/C or Unstable A pattern at 12 months, and three appeared depressed at 18 months. Three of the children were classified or force_classified B at some point. Again, the distinction among different subgroups of disorganized attachments did not appear especially revealing. Indeed, analyses described in the next section indicate that on some measures, the mothers of children whose attachments were considered only borderline disorganized ("D_signs") appeared most problematic.

Classification Scheme Versus Disorganization Level

The final set of analyses, partly redundant with the ANOVAs already conducted, demonstrated the importance of taking into account both (1) the style of the attachment pattern over time (forced classification) and (2) the amount of disorganization present, particularly by 18 months. These two pieces of data accounted for unique variance in data on the children, mothers, and family circumstances across the infancy period. The non_overlapping nature of their contributions is evident in analyses using forced classifications (A,B,C) versus level of disorganization (none,D_signs,D classification) at 12 months and 18 months as independent variables (along with their interaction) in separate analyses of variance to account for variation in maternal and infant data. Results appear in Table 9. With the exception of the demographic risk score, maternal reported anger control, and change in reported infant difficultness, either forced classifications or disorganization resulted in a significant model. However, there was no factor for which bothschemes demonstrated significant effects. When one scheme demonstrated a pattern of group differences, the other did not.

Thus, forced classifications at one or both ages explained differences on scores for maternal depressive symptoms ("C's" highest), anger control at 18 months only ("A's" lowest), relationship satisfaction ("A's" highest, "C's" lowest), and observed infant difficultness ("C's" highest). In previous analyses of caregiving (Vondra, Shaw, & Kevenides, 1995), mothers of infants with resistant (force classified) attachments at 12 months scored highest on a composite score of unresponsiveness to their infants' signals at the same assessment. Those having avoidant relationships with their infants scored highest on a composite of controlling behavior, and those having secure relationships scored highest on sensitive care. Level of disorganization, in contrast, explained differences on scores for hostile, suspicious personality ("D_Signs" highest) and for disruptive events and influences ("D's" highest). In the few instances where differences in 12_month maternal interactive behavior exist, it is the mothers of infants with only a moderate level of disorganization ("D_Signs") who looked worst on individual indices of care. Again, disorganization was also related to child maltreatment in the small number of known cases in this sample. Obviously, there is an association between relationship insecurity and maternal negative affectivity in general, but the form and expression differed considerably across the different patterns and quality of insecurity. Because these analyses to some extent repeat the ANOVAs conducted on subgroups, there is likely some inflation of alpha in reporting results. The pattern, however, is striking and deserves both replication and consideration.

The fact that either one or the other scheme provided significant discriminative power supports the original subgroup analysis, which in many ways represents an effort to capitalize both on the style and presence of attachment organization. By using both sets of information, we appear to have gained in our understanding of how individual and/or contextual factors can differ systematically with qualitative differences in the parent_infant relationship.

Summary and Conclusions

Results of this investigation support the need to consider both traditional patterns of attachment (A,B,C) and level of disorganization or atypicality as each change over time. Change in patterns of attachment was at least as common as stability over a six_month period for this sample of infants from socioeconomically disadvantaged families. On the one hand, a likely circumstance receiving little attention in the empirical literature is that attachment classification and change in classification may well reflect "error" variance associated, for example, with infant illness or fatigue and temporary disruptive family events, as well as coder unreliability. The somewhat low rate of stability in this and other disadvantaged samples could reflect greater coder unreliability in scoring infant adaptations to social class differences in parenting style, in stressful home experiences, and in changing social circumstances. The higher rate of stress and mental health problems in lower_income families, however, could also explain "lawful" discontinuities in attachment classification. Collaboration across laboratories in double_scoring of attachment videotapes is one strategy for studying these issues, as are social class comparisons of Strange Situation behavior conducted within a laboratory.

At the same time, attachment instability appears to be more common than originally thought, even among white, middle_class samples (Belsky, Campbell, et al., 1996). It seems clear from this and other research on attachment stability and change across infancy that there is a systematic component to change that covaries with individual maternal and infant factors, as well as with family circumstances. This systematic component involves meaningful variation both in the type of organized pattern discerned and in the degree of disorganization apparent over time, each of which have unique contextual correlates.

Organized insecurity was associated with maternal reported depressive symptoms, relationship dissatisfaction, and observed infant difficultness. Disorganized patterns and indices were associated with maternal reported aggressiveness and suspiciousness, with disruptions in or instability of family life and, anecdotally, with child maltreatment. Within the group of disorganized attachments, distinctions between forced classifications (secure, not secure) and between subtypes (A/C, Unstable A), though limited by small numbers, were notably unrevealing. More important were the boundaries of what constituted a "disorganized" attachment. By studying those children whose attachments appeared borderline disorganized ("D_signs"), and by including in the disorganized classification children with the Unstable A pattern, patterns of association emerged with maternal report and observer ratings. The inconsistency between these findings and results reported by Lyons_Ruth (this volume) may reflect sampling fluctuation, since both are based on small samples. They may also point to differences across laboratories in criteria for D classification. Viewing disorganization along a continuum of severity seems a fruitful approach at the present time, while distinctions among children showing atypical attachment patterns receive further examination.

Results replicated earlier findings that insecurity by 18 months is associated with indicators of maternal negative affect, but not with earlier maternal behavior that was unresponsive to, controlling of, or angry in relation to the infant. Self_reported anger distinguished mothers of infants with avoidant attachments, relationship dissatisfaction, mothers of resistant infants, and perceived hostility in self and others, mothers of infants with stable, disorganized attachments. Mothers in the resistant attachment group were at noticeably lower demographic risk (none were teen mothers, high_school dropouts, of minority race, or without a male partner) than were mothers in all but the stable secure group, but their infants expressed the most negative affect of any group at 12 months and their intimate relationships caused these mothers the most dissatisfaction.

This study was unique in being able to compare circumstances and behavior of mothers whose infants became secure or disorganized by 18 months, versus those whose infants showed a stable pattern of security or disorganization from 12 to 18 months. Mothers of infants who became secure at 18 months may have been less prepared psychologically for childrearing than mothers in the stable secure group in that they were often young and usually dealing with a firstborn child. Mothers of infants who became disorganized at 18 months, like mothers in the secure groups, scored lower on the personality complex of aggressiveness and suspiciousness (perceived hostility) that characterized mothers of infants showing stable disorganization, but highest of all groups on intervening, disruptive life events that they often did not see as having an effect on their child. Both change groups rated themselves higher on anger control than did mothers in the avoidant group. Across maternal report measures, there was an inconsistency of negative experiences and negative affect in the disorganized group that brought to mind Lyons_Ruth et al.'s (this volume) affectively contradictory maternal behavior.

Future Research on Atypical Attachment

Just as different attachments, including different forms of atypical attachments exist and either have proven or may prove developmentally informative (Crittenden, this volume; Lyons_Ruth et al., this volume), changes in attachment exist and appear to be ecologically informative. Work remains to be done, however, linking patterns of maternal reported negative affect and defendence to disorganized attachments in infancy (and to controlling, A/C, or AD attachments in the preschool period, see Crittenden, this volume; Solomon, George, & de Jong, 1995). Very little research tests the role of maternal interactive behavior as a possible mediator of maternal defendence and atypical attachment, although preliminary evidence indicates that a fruitful avenue is to examine frightened, frightening, and/or disorganizing behavior on the part of mothers Lyons_Ruth, this volume; Schuengel et al., in press). But there is an important missing piece, which is the infant's experience in the home. Family violence (spouse abuse, sibling abuse), parental substance abuse, traumatic parent_child separations, frequent disruptions in caretaking arrangements, and other extreme experiences are rarely assessed directly and probably are captured only partly in measures of maternal interactive behavior. But it is these more extreme experiences that may distinguish homelife for mothers with unusual defendence against negative affect and for infants and children with atypical attachments.

Clearly, there is ecological significance both in the form of secure and insecure attachments and in the presence of behaviors indicative of disorganization. Both aspects of infant_mother attachment covaried with infant, maternal, and familial characteristics. Until more data become available that test these distinctions, we can only offer hypotheses regarding the meaning of these two dimensions of early infant_mother attachment. It is consistent with the growing literature on adult attachment (e.g., Ainsworth & Eichberg, 1991; Cassidy & Kobak, 1988; Grossmann & Grossmann, 1991) to suggest that parental styles of acknowledging and dealing with negative affect and infant temperamental styles distinguish early attachments according to traditional, or "forced," classifications, but that more serious indicators of parental pathology (clinical depression, personality problems, childhood experiences of parental loss or maltreatment) and of unstable, chaotic, abusive child experiences distinguish disorganized or atypical attachment behaviors or patterns in infancy. If one point seems clear at this stage, it is that atypical or disorganized attachment patterns in infancy and early childhood do appear to be an important link between variations in early experiences, behavior, and functioning that may be considered "normative," and are studied under the rubric of developmental psychology, and those that are more extreme or deviant and may be considered "non_normative," more often studied under the rubric of developmental psychopathology. Whether atypical or disorganized attachment patterns will prove to be both an empirically and clinically useful prognostic indicator of emerging psychopathology remains to be seen.