It is well established that maternal depression, whether defined in terms of clinical diagnosis or self-reported depressive symptoms, is considered a risk factor for a range of negative cognitive and socioemotional outcomes in young children (e.g., Cummings & Davies, 1994; Downey & Coyne, 1990; Gelfand & Teti, 1990). Studies have documented more negative affect and poorer affect regulation, more behavior problems, less cooperation, and poorer cognitive and language functioning in children whose mothers are clinically depressed (e.g., Cohn & Campbell, 1992; DeMulder & Radke–Yarrow, 1991; Murray, 1992; Zahn–Waxler, Iannotti, Cummings, & Denham, 1990) or report high levels of depressive symptoms (e.g., NICHD Early Child Care Research Network [ECCRN], 1999a). Moreover, the effects of maternal depressed mood and symptoms on development may be especially evident during infancy, toddlerhood, and the preschool years because young children are more dependent upon stimulation, nurturance, and scaffolding from primary caregivers (Beardslee, Bemporad, Keller, & Klerman, 1983). Because of the importance of maternal responsiveness to children's optimal socioemotional development, it is not surprising that maternal depression has also been considered a risk factor for insecure attachment (Martins & Gaffan, 2000). In the present study, we examine the relations between maternal depressive symptoms and preschool attachment security in a large sample of children followed from birth.

According to attachment theory (Bowlby, 1969; Carlson & Sroufe, 1995) and empirical findings (DeWolff & van IJzendoorn, 1997), mothers who are more sensitive and responsive are more likely to have children who are securely attached. Although maternal sensitivity accounts for only a small proportion of the variance in attachment security, with an average effect size across 30 studies of r = .22 (DeWolff & van IJzendoorn, 1997), the link between maternal sensitivity and attachment security is robust. From an attachment theory perspective, maternal depression may well represent a prime example of an “experiment of nature” (Cicchetti, 1989) because the very symptoms that define depression, for example, sad mood, loss of interest, disengagement, fatigue, and irritability, should be associated with lower maternal sensitivity, which in turn would be expected to predict lower rates of attachment security in young children. This is because the symptoms of maternal depression, whether or not they reach diagnosable levels, may be experienced by the young child as unresponsive, inconsistent, unavailable, or rejecting care (Carlson & Sroufe, 1995; Cummings, Davies, & Campbell, 2000). Consistent with this view, a growing literature indicates that women who report elevated symptoms of depression are less positive, more negative, and less engaged with their young children (e.g., Campbell, Cohn, & Meyers, 1995; DeMulder & Radke–Yarrow, 1991; Field, 1992; NICHD ECCRN, 1999a; Teti & Gelfand, 1991) across infancy, toddlerhood, and the preschool years. Thus, it is reasonable to expect that the disengagement and lack of responsiveness characteristic of many depressed mothers will be reflected in higher rates of insecurity in their children in comparison to children of nondepressed mothers. Therefore, we examined rates of attachment insecurity as a function of maternal depressive symptoms and then asked whether maternal sensitivity mediated any associations between maternal depressive symptoms and young children's attachment security. Alternatively, the combination of both high levels of depressive symptoms and lower maternal sensitivity may be associated with increased rates of insecurity, consistent with a dual risk model wherein depressive symptoms and low levels of sensitivity have synergistic effects on preschool children's attachment security.

Studies of attachment using the Ainsworth Strange Situation or variations on that paradigm (Ainsworth, Blehar, Waters, & Wall, 1978; Cassidy, Marvin, & the MacArthur Working Group on Attachment, 1992; Crittenden, 1992) have classified insecure infants, toddlers, and preschool children into avoidant, resistant, and disorganized subgroups. Preschool children considered avoidant show little affect, either positive or negative, during separation or reunion episodes (Cassidy et al., 1992). Insecure-resistant preschool children (C) tend to show ambivalent approach behavior, and they are not comforted by their mother's return. In the toddler or preschool system (Cassidy et al., 1992), the D category includes children who show controlling behavior or a mix of insecure strategies. After separation from mother in the laboratory, 2- and 3-year-olds classified as D may try to take over, either by being solicitous, or by being punitive and angry toward mother. Other preschoolers classified as “insecure other” (D) may show combinations of strategies during reunion such as ambivalence and avoidance or avoidance mixed with controlling behavior. In contrast, children who are secure (B) use their mother as a base for exploration and as a source of comfort and support throughout infancy and early childhood.

The behaviors that characterize insecurity in preschoolers of depressed mothers may reflect children's varying reactions to separation and reunion, given that their mothers are presumably sometimes less psychologically available and/or more negative. For example, maternal irritability and intrusive behavior may be associated with avoidance as children learn that attachment bids may provoke rejection, and thus it is better not to be too demanding (Isabella & Belsky, 1991; Main, 1981). Inconsistency in caregiving may be associated with resistant and clingy behavior, as children learn that heightened emotionality sometimes succeeds in eliciting a response from a preoccupied caregiver (Cassidy & Berlin, 1994). The controlling behavior characteristic of some preschool children with D attachments also may be a reaction to the passivity and lack of engagement that characterizes some depressed mothers, leading the child to take charge of the interaction (Teti & Gelfand, 1997). We are not suggesting that these are conscious strategies but are instead potentially adaptive behaviors that have emerged over the course of interaction with a sometimes less available mother, because demanding behavior may be a successful way of obtaining maternal attention (Greenberg, Speltz, & DeKlyen, 1993).

The theoretical link between maternal depression and attachment security in young children seems obvious, but studies examining this association have shown equivocal results. Martins and Gaffan (2000) conducted a meta-analysis on seven studies of child–mother attachment in samples of women with clinical diagnoses of unipolar depression. Four studies found that children of depressed mothers were more likely to be insecure (Hipwell, Goossens, Melhuish, & Kumar, 2000; Murray, 1992; Seifer, Sameroff, Dickstein, Keitner, Miller, Rasmussen, & Hayden, 1996; Teti, Gelfand, Messinger, & Isabella, 1995) and three studies reported no such association (Campbell & Cohn, 1997; Dawson, Klinger, Panagiotides, Spieker, & Frey, 1992; DeMulder & Radke–Yarrow, 1991). Children in these relatively small samples ranged in age from 12 to 52 months, and the timing of depression also varied. Martins and Gaffan (2000) concluded that “both the insecure-avoidant (A) and disorganized (D) categories increase to a marginally significant extent among infants of depressed mothers,” with the relation between maternal depression and D classifications somewhat more consistent. They suggested that the timing of the assessments, co-occurring risk factors, and maternal behavior may have accounted for inconsistent findings. Results also appeared to be more consistent when toddlers and preschoolers were studied and when co-occurring risk was in evidence.

Another meta-analysis of the link between depression and Strange Situation attachment security based on 15 studies, some of which overlapped with the Martins and Gaffan meta-analysis, found an overall effect size of r = .18 for the association between depression and insecure attachment (Atkinson, Paglia, Coolbear, Niccols, Parker, & Guger, 2000). However, the effect size was quite heterogeneous across studies and was stronger in clinical samples (r = .27) and weaker in nonclinical samples (r = .09). This may be because some of the women in studies of community samples are showing only temporary distress. Thus, studies that include multiple assessments of depressive symptoms may be more likely to identify community women with more serious problems (NICHD ECCRN, 1999a). In yet another meta-analysis, van IJzendoorn, Schuengel, and Bakermans–Kranenburg (1999) examined the link between maternal depression and D attachment and also found inconsistent results. Atkinson et al. (2000) report no difference in effect sizes for studies using the D classification, compared to those that only used A, B, and C classifications.

Cicchetti, Rogosch, and Toth (1998) used the attachment Q-set to assess attachment in toddlers of depressed mothers. They found much higher rates of mother-rated insecurity in children whose mothers met diagnostic criteria for depression than in offspring of control mothers. This study includes a much larger sample of depressed women, but the use of the attachment Q-set instead of a laboratory based observation of separation and reunion behavior makes it difficult to compare these results to other studies of attachment security and maternal depression.

These inconsistent results highlight problems comparing studies that have examined attachment in children of different ages, using different classification systems (Ainsworth, MacArthur, Crittenden, Q-set). In addition, differences in the course, severity, and timing of maternal depression may need to be considered. Campbell, Cohn, and Meyers (1995) studied women with depression that was diagnosed in the postpartum period and found that the chronicity of maternal depression was related to maternal sensitivity and responsiveness across assessments at 2, 4, and 6 months during feeding and play. However, chronicity through 6 months was not related to attachment security (Campbell & Cohn, 1997). This may be because most of the depressions had remitted by 6 months postpartum, or because, in general, this was a low risk sample. In addition, women with depression vary widely in their behavior with their children (Campbell & Cohn, 1997), so that group designation is not automatically a proxy for inconsistent or disengaged parenting. Differences in the proximal features of maternal behavior may partly explain discrepant findings. Thus, studies with large enough samples to examine the course and timing of maternal depressive symptoms, maternal behavior, and patterns of insecurity are needed.

As already suggested, the links between attachment security/insecurity and depression may be at least partly a function of other aspects of the mother–child relationship, especially maternal sensitivity and responsiveness. Using the same data set as the current report, the NICHD ECCRN (1999a) examined how observed maternal sensitivity during a semistructured play interaction varied from 6 to 36 months in concert with self-reported symptoms of depression by comparing groups defined as never depressed, sometimes depressed, and chronically depressed. As expected, women who never reported elevated symptom levels were the most sensitive. Women reported symptoms sometimes and chronically differed from each other in patterns of change, with women who reported chronic symptoms appearing especially insensitive at the 24-month assessment, possibly because their toddlers' needs for autonomy at this developmental stage made parenting particularly challenging for them. In addition, women who reported depressive symptoms, especially more chronic symptoms, were more likely to have preschoolers who performed poorly on measures of school readiness and language development and who were rated as having more behavior problems at 36 months. However, the effects of maternal depression on cognitive and language outcomes were partially accounted for by levels of maternal sensitivity, although these effects were modest.

Maternal sensitivity also moderated the association between maternal depressive symptoms and developmental outcomes. The combination of depressive symptoms and less sensitive maternal behavior predicted lower expressive language abilities in children at 36 months and maternal ratings of more problem behavior and less cooperation. Conversely, maternal sensitivity acted as a buffer when mothers reported depressive symptoms. Not all women who report depressive symptoms are disengaged from their children, and the effects of maternal depressed mood and symptoms on the quality of the mother–child relationship and on child outcomes may vary with more proximal maternal behavior toward the child. Thus, mothers who report symptoms of depression, but who are also sensitive and responsive, may be no more likely to have children who are classified as insecure than are mothers not reporting depressive symptoms. In contrast, the dual risk of depressive symptoms, especially more prolonged symptoms and maternal insensitivity, should be associated with higher rates of insecurity. The combination of chronic symptoms and high levels of insensitivity are likely to expose young children to more continuous experiences with rejection and negative affect, whereas when mothers are able to be sensitive despite their reports of depressive symptoms, even chronic symptoms, their children are less likely to be regularly exposed to the parenting sequelae of these symptoms.

In addition to the NICHD study, data from DeMulder and Radke–Yarrow (1991) are consistent with the hypothesis that maternal behavior during mother–child interaction may moderate the links between depressive symptoms and insecurity. They found that depressed mothers with insecure children were observed to show more sadness and irritability with their toddlers when observed in a naturalistic apartment setting. Their toddlers in turn were less affectionate with their depressed mothers. However, this pattern of mother–child interaction did not characterize the dyads with depressed mothers and securely attached toddlers or the nondepressed control dyads. These findings underscore the importance of studying individual differences in maternal behavior among symptomatic women in order to better explain individual differences in children's functioning. Thus, we will examine maternal sensitivity as a moderator of the association between insecure attachment and depressive symptoms.

In addition to individual differences in maternal behavior among mothers reporting depressive symptoms, it is important to consider variations in the course and timing of these symptoms. Some women show brief episodes of depression that do not appear to be reflected in less involved maternal behavior, whereas more prolonged depression is associated in general with less positive and sensitive engagement (Campbell & Cohn, 1997; Campbell et al., 1995; NICHD ECCRN, 1999a). In addition, the timing of the depression during the child's life may be relevant. Few studies address this issue and they are at best inconclusive, given the problems with small samples (e.g., Campbell et al., 1995; DeMulder & Radke–Yarrow, 1991; Murray, 1992) and retrospective data on the timing of symptom onset.

When studying attachment security, however, the timing of the depression may be especially important. For example, depression over the course of the first year when the attachment relationship is developing may be more deleterious to the infant's development of a sense of felt security than depression that develops during the toddler's second year. On the other hand, depressive symptoms may be of greater concern if they interfere with maternal engagement and gentle guidance during the child's second and third years, as the “goal corrected partnership” becomes consolidated (Bowlby, 1969; Cicchetti, Cummings, Greenberg, & Marvin, 1990) and a sense of autonomy develops. Maternal depressive symptoms experienced during this developmental period may be reflected in lower rates of security at 36 months. Generally, more short-lived depressions, regardless of timing, may be less salient than depressions that are more prolonged, more frequent, or more unpredictable. Both chronic depressive symptoms and symptoms that wax and wane over the course of the child's development may have a more negative effect on the quality of the mother–child relationship than symptoms that cover a briefer period of young children's lives. Although this issue has not been examined to our knowledge, the meta-analysis by Martins and Gaffan (2000) did hint at the possibility that depression that is evident in toddlerhood and the early preschool period may be more strongly associated with insecurity, especially when other indicators of family risk are also evident, which is consistent with the findings of Cicchetti et al. (1998). The availability of a large data set with repeated assessments of maternal depressive symptoms allowed us to explore this possibility.

In summary, although meta-analyses find a significant association between maternal depression and attachment security, there is heterogeneity across studies. This may be due to methodological variations, including sample composition, and the definition and severity of depression. It may also be because maternal sensitivity varies among mothers reporting depressive symptoms and serves as a mediator or moderator of the effects of depressive symptoms on attachment, a possibility largely overlooked in prior research. Further, the course and timing of maternal depressive symptoms may affect associations between depression and attachment, and this possibility has not yet been examined systematically either. The NICHD Study of Early Child Care data set permits us to address these issues because of the large sample and the repeated assessments of both maternal depressive symptoms and maternal sensitivity over the first 3 years of life. The 36-month time point was chosen because it permits us to examine the course of depression symptoms beyond infancy and toddlerhood and to assess attachment at a time when children are becoming more independent and when the attachment relationship as a goal-corrected partnership should be well established (Cicchetti et al., 1990).

Using data from the NICHD Study of Early Child Care, we examined the following questions:

1. Are children whose mothers report depressive symptoms at any time over the course of their first 3 years of life more likely to be insecurely attached when observed in the laboratory with their mothers in a modified Strange Situation at 36 months than children whose mothers do not report elevated depressive symptoms?
2. Does the course and timing of maternal depressive symptoms over infancy and early childhood relate systematically to attachment security at 36 months? Comparisons between women reporting no depressive symptoms and those reporting early, late, intermittent, or chronic patterns of depressive symptoms will allow us to address this question.
3. Are relations between maternal depressive symptoms and attachment security mediated by maternal sensitivity? That is, does maternal sensitivity account for the association between depressive symptoms and patterns of insecurity?
4. Does maternal sensitivity moderate the association between maternal depressive symptoms and attachment security? Are children whose mothers report depressive symptoms but who are also sensitive when interacting with them less likely to be insecurely attached than children of depressed mothers who are insensitive? In particular, do maternal depressive symptoms, especially chronic or intermittent symptoms, and insensitivity work together to predict especially high rates of insecurity?

Results

Preliminary analyses

As discussed earlier, the validity of the MacArthur modified Strange Situation and its associated scoring criteria was established in a prior publication from the NICHD ECCRN (2001). Before we examined the associations between maternal depression scores and attachment security, we asked whether observed maternal sensitivity at each age differentiated between children who were secure and insecure (A, C, and D combined) at 36 months. Analyses of variance (ANOVAs) were used to compare the secure and insecure groups on the maternal sensitivity composite score obtained from observations during free play at 6, 15, 24, and 36 months. At each age preschoolers who were securely attached at 36 months had mothers who were rated as significantly more sensitive and responsive during play than mothers whose children were later rated as insecure at 6 months, Ms and SD = 0.06 (0.97) and −0.07 (1.03), F (1, 1062) = 4.23, p = .04; 15 months, Ms and SD = 0.11 (0.94) and −0.09 (1.04), F (1, 1066) = 10.30, p < .001; 24 months, Ms and SDs = 0.18 (0.90) and −0.22 (1.07), F (1, 1054) = 41.26, p < .001; and 36 months, Ms and SD = 0.15 (0.88) and −0.18 (1.09), F (1, 1067) = 30.54, p < .001. As reported in the earlier paper (NICHD ECCRN, 2001) boys were slightly more likely to be secure than girls (64.6 vs. 59.6%), χ² (1) = 2.96, p = .09. When all four attachment groups were examined, more girls than boys were classified as D (19.3 vs. 12.9%), χ² (3) = 9.13, p < .03.

36-Month attachment security and maternal depressive symptoms

Mothers were considered to be showing depressive symptoms if they scored at or above 16 on the CES-D at any assessment from 1 to 36 months. A dichotomous depression score was generated for this analysis (ever/never). Attachment was examined using the four-way classification of insecure avoidant (A), secure (B), insecure-resistant (C), and insecure-controlling/insecure-other (D). Data were analyzed in a 2 × 4 contingency table using the EXACON module in SLEIPNER (Bergman & El-Khouri, 1987, 1998), which provides the binomial probability that the observed and expected frequencies in each cell differ significantly from one another. The overall chi square was significant, χ² (3) = 15.51, p < .01. Examination of individual cells revealed that children of women who reported elevated depressive symptoms at least once were more likely to be classified as D attachments at 36 months (p < .01). This contrasts with children whose mothers never reported depression, who were less likely to be classified as Ds (p < .01) than would be expected by chance.

The course and timing of maternal depressive symptoms and attachment security at 36 months

Because symptoms of depression may be chronic in some women, short lived in others, and episodic in still others, it was important to examine how the course and timing of depression symptoms in women over the first 36 months of their child's life was related to attachment security. Thus, we classified the 1,077 women with at least four depression ratings and an attachment observation at 36 months into five groups on the basis of their symptom ratings at each time point: never depressed women (n = 594, 55.2%) received scores below 16 on the CES-D at each assessment point. Women in the early depression group (n = 198, 18.4%) reported elevated symptoms at the 1-, 6-, and/or 15-month assessment, but not thereafter. (Women with elevated symptoms in the postpartum period only and those with elevated symptoms at 6 and/or 15 months did not differ in maternal sensitivity or in rates of secure versus insecure attachment, so these two groups were combined to decrease the number of subgroups.) Women who did not report depressive symptoms until the 24- and/or 36-month assessment were classified into the late depression group (n = 98, 9.1%). Women who reported elevated symptoms of depression at least twice, separated by a period of lower scores (i.e., <16), were characterized as showing intermittent depression (n = 100, 9.3%). Finally, a group of chronically depressed women (n = 87, 8.1%) reported elevated levels of depressive symptoms at least three out of the five times, with the majority reporting elevated symptoms either at each assessment (n = 40) or at four out of five assessments (n = 39); another eight women reported elevated symptoms at 15, 24, and 36 months, reflecting chronic symptoms across the second and third years of their child's life.

For women missing one depression score, the mean of the two adjacent scores was interpolated. This was done for 66 (6.1%) women who fell into the following groups: never (4%), early (5.1%), late (12.2%), intermittent (4%), and chronic (18.4%). Although women in the chronically depressed group were the most likely to be missing one CES-D score, their scores were elevated at the four assessments with complete data; the means for each woman in this group on the four available scores ranged from 17.7 to 36.3. The average depression scores across all assessment points for each group were as follows: never (M = 5.01, SD = 2.66), early (M = 11.64, SD = 3.31), late (M = 11.53, SD = 3.16), intermittent (M = 15.81, SD = 3.08), and chronic (M = 23.70, SD = 5.07). Thus, the chronic depression group had consistently high levels of self-reported depression symptoms across their children's lives, whereas the average depression scores over time for the other groups were substantially lower.

Women in the five depressive symptom groups were compared on demographic variables. In general, women who never reported depressive symptoms were significantly more likely to be partnered, and to have a higher educational level and income than women in the groups reporting depressive symptoms; women reporting chronic symptoms were generally lowest on demographic indicators and they differed from the early and late depression symptom groups on education and income. Women in the intermittent and chronic groups were also somewhat younger on average than women in the other three groups. Infant gender did not vary with depression symptom group, χ² (4) = 1.25, ns. These data are displayed in Table 1.

Demographic variables and average maternal sensitivity by the five depression groups

The maternal sensitivity scores were converted to standard scores and averaged across the 6-, 15-, 24-, and 36-month observations. Consistent with a prior report (NICHD ECCRN, 1999a), women in the never depressed group were most sensitive. Women in the intermittent and chronic depressive symptom groups were the least sensitive. A one-way ANOVA indicated a significant difference across groups, F (4, 1072) = 32.65, p < .0001. Post hoc tests indicated that women in the never depressed group were significantly more sensitive than women in the four depression symptom groups (p < .001). Women who reported chronic symptoms were significantly less sensitive than women in the early and late groups (p < .01) who did not differ from each other.

When attachment security was examined as a function of depression group using the EXACON procedure, the association between depression and attachment was significant, χ² (12) = 32.49, p < .001. An examination of discrepancies between observed and expected frequencies within individual cells in the contingency table revealed that women reporting intermittent symptoms of depression had young children who were more likely than expected to be classified as C (resistant, p < .03) and less likely to be classified as B (secure, p < .04). Women whose symptoms were chronic were more likely to have children who were classified as D (insecure-controlling or insecure-other, p < .01). Surprisingly, women in the early depression group were less likely than expected to have children who were classified as A (avoidant, p < .05). However, the small number of avoidant children makes interpretation of this finding difficult, as only 4.6% of the entire sample was classified as avoidant. In addition, as reported earlier, women who never reported depressive symptoms were less likely to have children who were classified as D (p < .01) than would be expected by chance. In order to compare the effect size for each depression group with the meta-analysis reported by Atkinson et al. (2000), four 2 × 2 contingency tables, depression group (never vs. early, late, intermittent, or chronic) by attachment (secure–insecure), were analyzed. The results indicate a small, but significant association with insecurity for the intermittent (r = .12) and chronic (r = .07) groups. Data on attachment security by depression group are displayed in Table 2.

Depression chronicity by attachment security at 36 months

In summary, then, the course and timing of depressive symptoms were related both to maternal sensitivity and attachment security at 36 months. Women who reported depressive symptoms were less sensitive with their young children than were women who never reported feeling depressed; women reporting chronic symptoms were the least sensitive in independent observations across the 6–36 month period. Women who reported either chronic or intermittent symptoms were more likely than expected by chance to have insecurely attached children who showed either insecure-controlling/insecure-other, or resistant strategies. In contrast, children of women reporting only early or late symptoms did not show higher than expected levels of insecurity at 36 months.

Sensitivity as a mediator and moderator of depression effects

Given these patterns of association between the course of maternal depressive symptoms and 36-month attachment, we next examined whether third variables were accounting for these patterns and then whether maternal sensitivity mediated or moderated the effect of depression symptom group on attachment security. To this end, a multinomial logistic regression (Hosmer & Lemeshow, 1989) was conducted to predict the four preschool attachment classifications. Demographic variables were entered first as control variables (child gender, mother's education, income to needs ratio, and proportion of time mother with partner) followed by depression symptom group (never, early, late, intermittent, and chronic). In these analyses, each insecure attachment group (A, C, and D) was compared with the secure attachment group (B) as the reference. Depression groups were compared with the never group. These results are summarized in Table 3.

Multinomial regression predicting attachment classification at 36 months

The main effects of depression symptom group on attachment classification

The first overall model (Model I), which involved testing for demographic and depression group effects, was significant, χ² (24) = 66.41, p < .0001. Likelihood ratio tests indicated that maternal education, partner status, child gender, and maternal depression group were significant predictors of attachment classification. After controlling for demographic variables, women who reported symptoms of depression early in their child's life were less likely to have preschoolers who were avoidant (A) than women who never reported elevated symptoms. Women who reported intermittent depressive symptoms were more likely to have children who were classified as either resistant (C) or insecure-controlling/insecure-other (D) than never depressed women; women with chronic symptoms also were more likely to have children classified as D than were women never reporting symptoms. These findings are consistent with the EXACON analysis, but control for potentially confounding demographic variables.

The relationship between maternal sensitivity and the depression–attachment link

Model II included maternal sensitivity averaged over 6, 15, 24, and 36 months, to determine whether sensitivity mediated associations between maternal depression symptom patterns and attachment security. Again the overall model was significant, χ² (27) = 88.74, p < .0001. Likelihood ratio tests indicated that average sensitivity, child gender, and depression group were significant predictors of attachment classification in the overall model. Thus, with maternal sensitivity included in the model, maternal education and partner status were no longer significant predictors, but depression group continued to be significant. It was not surprising that maternal sensitivity was a strong predictor of attachment classifications, χ² (3) = 22.32, p < .01.

In looking at individual contrasts, with sensitivity included in the model the link between intermittent symptoms and resistant (C) attachment dropped to nonsignificance; that is depression group (never vs. intermittent) was not a significant predictor, although lower sensitivity predicted higher rates of resistance. However, the evidence for mediation was weak because the odds ratio linking the depression group to resistant attachment only dropped from 1.72 to 1.60, a trivial effect.

The D attachment status was predicted by average sensitivity and child gender, with lower sensitivity and being a girl predicting higher rates of insecure-controlling/insecure-other attachment, which is consistent with the analyses reported earlier. With sensitivity in the model, the association between both intermittent and chronic depressive symptoms and D attachment was attenuated; mothers reporting intermittent symptoms were still significantly more likely than women never reporting symptoms to have children who were insecure D, whereas membership in the never and chronic depression symptom groups was no longer a significant predictor of a D classification. In looking at the strength of the mediation effect, it is noteworthy that the odds ratios showed only modest decreases (from 2.09 to 1.83 and from 2.18 to 1.79), suggesting weak evidence for mediation and underscoring the importance of considering patterns of depressive symptoms and maternal sensitivity simultaneously.

Interactions between patterns of maternal depressive symptoms and maternal sensitivity in predicting attachment classification

Model III examined whether maternal sensitivity moderated the association between depression symptom group and attachment security, on the assumption that the combination of low sensitivity and more chronic or intermittent depressive symptoms would increase the likelihood of insecurity, especially a D classification. Thus, the interaction between maternal sensitivity and depression symptom group was added to the model which remained significant, χ² (39) = 113.03, p < .0001. When the interaction terms were included, only child gender and the Depression Group × Sensitivity interaction were significant predictors of attachment classification. There were several significant interaction terms of interest: sensitivity by late versus never and intermittent versus never predicted resistant versus secure attachments (C vs. B); sensitivity by late versus never, intermittent versus never, and chronic versus never predicted D attachments (insecure-controlling/insecure-other) versus secure attachments (D vs. B).

To identify the locus of the significant moderator effects, maternal sensitivity was split at the median and the number of secure versus insecure attachments was examined as a function of the level of sensitivity and depression symptom group (never vs. relevant depression group). Specifically, 2 (B vs. C or D) × 4 (high sensitivity never, low sensitivity never, high sensitivity depression group, low sensitivity depression group) chi-square analyses were conducted using the EXACON procedure to pinpoint significant cells. Recall that it was predicted that women with high levels of depressive symptoms who also showed lower levels of sensitivity would be most likely to have preschoolers showing an insecure (C or D) attachment. When interactions between depression symptom group and high or low sensitivity were examined as predictors of resistant versus secure attachments using EXACON, it was determined that women with intermittent symptoms of depression who were also insensitive were especially likely to have preschoolers who were resistant (p < .001) and less likely to have preschoolers who were secure (p < .03). Indeed, as can be seen in Table 2, 26 women with intermittent depression had preschoolers classified as resistant; 22 of these women were also low in sensitivity. A similar pattern was evident when women reported being depressed only later (24 and/or 36 months), but none of the cells showed significant differences between observed and expected frequencies of B or C attachments, presumably because of the small number of late depressed women with children who were resistant (15 were resistant; of these, 12 mothers were in the low sensitivity group).

Similar analyses were conducted examining the interactions between depression group and sensitivity in the prediction of D versus B attachments and a similar dual risk pattern was in evidence (see Figure 1). In each case, when mothers reported depressive symptoms and were rated as lower in sensitivity, their preschoolers were more likely to show the D pattern of attachment. Women who reported being intermittently depressed and who were also insensitive were especially likely to have preschoolers who were classified as D (p < .003); of the 21 D offspring with intermittently depressed mothers, 16 had mothers who were also insensitive when observed interacting with them across the first 3 years of life. Women whose depression was only reported in the toddler/preschool period showed a similar picture (17/21, p < .01). Finally, children whose mothers reported chronic symptoms were especially likely to show D attachments (24 or 27.6%) and the majority of these children (21) also had mothers rated insensitive (p < .001). Further, children with chronically depressed and less sensitive mothers were less likely than expected by chance to be secure (p < .04).

Interactions between depression group and maternal sensitivity in predicting insecure (D) attachment.

In summary, when women reported late, chronic, or intermittent symptoms of depression, and were also observed to be low in maternal sensitivity, their 3-year-olds were more likely to be insecurely attached (C or D), whereas these same patterns of maternal depressive symptoms coupled with high maternal sensitivity did not appear to place the child at greater risk of attachment insecurity, as is evident from Figure 1.