Attachment and affect between mothers with depression and their children: Longitudinal outcomes of Child Parent Psychotherapy

Abstract

The present investigation examined the longitudinal effects of Child-Parent Psychotherapy (CPP) for toddlers and their mothers with depression on: a) maternal affective expression, b) child affective expression, and c) mother-child cohesion. Mothers with depression (M_age = 31.7 years; 92.8% White, 3.5% Black, 2.1% Hispanic, 2.3% other) and their toddlers were randomized to receive CPP (DI; n = 66) or to a control group (DC; n = 64). Mothers without depression and their toddlers (NC; n = 68) were recruited as an additional comparison group. Dyads were assessed at baseline (T1; 20 months old), post-intervention (T2; 36 months old), and follow-up (T3; 9 years old). Data from a mother-child conflict task was coded as a measure of observed outcome variables. Change in post-intervention attachment security assessed via the Strange Situation was evaluated as a mediator between intervention condition and maternal and child affective expression and dyadic cohesion at T3. Change to secure attachment post-intervention significantly mediated the association between intervention condition and T3 maternal warmth and child anger/problem behavior. Toddlers of mothers with depression who received CPP showed higher rates of change to secure attachment compared to those in both the DC and NC groups. Dyads who changed to secure attachment at T2 displayed higher levels of maternal warmth at T3 and lower levels of child anger and problem behavior at T3. Implications for the use of CPP as a preventive intervention and the importance of attachment as a mediator of long-term outcomes are discussed.

According to the World Health Organization (2008), major depressive disorder (MDD) is ranked as the third most disabling disorder worldwide and the leading cause of disease-related disability among women. Although it has largely been thought that mothers are at the greatest risk for depression during the peripartum period, newer evidence suggests that the first few years following childbirth may be of comparable significance (Ertel, Rich-Edwards, & Koenen, 2011). Woolhouse and colleagues (2015) found that almost 1 in 3 women reported depressive symptoms during the 4 years after the birth of their child, with the prevalence of maternal depression 14.5% at 4 years postpartum, compared to only 8.8% during pregnancy and 8.1% at 3 months postpartum.

With this high prevalence of maternal depression, there is also increased risk for impaired caregiving, which can impact the mental health and behavior of children of mothers with depression. Mothers with depression tend to be more disengaged with their children (Pelaez, Field, Pickens, & Hart, 2008) and less sensitive and responsive in their parenting skills (Hoffman & Drotar, 1991). As a result, children of mothers with depression are at increased risk for lower cognitive functioning (Cicchetti, Rogosch, & Toth, 2000), poor socioemotional adaptation (Murray et al., 1999), and ineffective emotion regulation (Maughan et al., 2007). Meta-analytic data have suggested that children of mothers with depression are at increased risk for internalizing problems, externalizing problems, and general psychopathology (Goodman et al., 2011). Therefore, it is important to determine how best to intervene with mothers with depression and their children in order to support positive developmental outcomes.

Maternal Depression and Attachment

As theorized by Bowlby (1969), children can experience the physical and emotional unavailability of a caregiver with depression as a form of loss that damages the formation of a secure attachment relationship. Children with unresponsive, insensitive, and rejecting caregiving during the early years of life show an increased incidence of insecure attachment (de Wolff & van IJzendoorn, 1997). Because the parenting behaviors associated with attachment security overlap with those that are impeded by depression, there is evidence to suggest that being raised by a parent with depression confers risk for the development of insecure attachment relationships (e.g., Campbell et al., 2004).

The attachment relationship is thought to lead to the development of representational models that influence how children interpret their own behavior and the behavior of others (Bowlby, 1982). In doing so, attachment serves as the foundation from which children begin to form a sense of self, learn to relate to others, and develop self-regulatory capacities (Sroufe, 2005). Therefore, insecure attachment poses a risk for a number of negative developmental outcomes, including internalizing behavior, externalizing behavior, and lower social competence with peers (Fearon et al., 2010; Groh et al., 2012; Groh et al., 2014).

Although some researchers maintain that these early representational models are retained over time, others believe that they are subject to modification on the basis of new experience (Fraley, 2002). These differing perspectives have led to mixed findings on the continuity of attachment throughout development. One key variable in the continuity of attachment is sensitive maternal support. In a sample of adolescents adopted in early childhood, consistently sensitive maternal support in early childhood and adolescence predicted continuity of secure attachment from 1 to 14 years, whereas change from insensitive to sensitive maternal support predicted change from insecure to secure attachment from early childhood to adolescence (Beijersbergen, Juffer, Bakermans-Kranenburg, van IJzendoorn, 2012). Considered in the context of early intervention, these findings suggest that if intervention-related increases in maternal sensitivity can be maintained throughout childhood, children may be more likely to achieve stability in secure attachment and enjoy the positive developmental outcomes that accompany secure attachment. Therefore, the benefits of early intervention may extend beyond early childhood and have implications for children’s overall developmental trajectories if effects on maternal behavior can be maintained.

Child-Parent Psychotherapy

One of the most empirically supported attachment interventions is Child-Parent Psychotherapy (CPP; Lieberman, Ghosh Ippen, & Van Horn, 2015). CPP is an attachment theory-informed intervention intended to enhance the quality of the parent-child relationship and foster secure attachment. Unlike interventions that utilize skills-based training for parents, the primary focus of CPP is on the relationship between parent and child, rather than on each as individuals.

CPP has contributed to a variety of positive child and caregiver outcomes. Among preschoolers exposed to marital violence and other traumatic stressors, those who received CPP showed fewer behavior problems and traumatic stress symptoms (Lieberman, Van Horn, & Ghosh Ippen, 2005) and sustained decreases in problem behavior 6-months post-intervention (Lieberman, Van Horn, & Ghosh Ippen, 2006). Compared to a psychoeducational parenting intervention, CPP was more effective in decreasing children’s negative representational models of themselves and their mothers, increasing their positive self-representations, and increasing their positive mother-child relationship expectations as measured by narrative story-stems (Toth, Maughan, Manly, Spagnola, & Cicchetti, 2002). In addition, CPP was associated with higher rates of secure attachment and lower rates of disorganized attachment 12 months post-intervention compared to the psychoeducational control intervention (Stronach, Toth, Rogosch, & Cicchetti, 2013). These findings were the first to demonstrate the superiority of CPP over alternative interventions for fostering sustained improvements in attachment security. With respect to caregiver outcomes, CPP has been shown to be effective in increasing levels of observed maternal empathy and goal-corrected partnership (i.e., greater eagerness and reciprocity to interact; lower mutual indifference, awkwardness, hesitancy, mixed messages, or anger; Lieberman, Weston, & Pawl, 1991).

CPP has also been specifically examined within the context of maternal depression. CPP is effective in protecting cognitive development and in increasing rates of secure attachment among children of mothers with depression (Cicchetti, Rogosch, & Toth, 2000; Cicchetti, Toth, & Rogosch, 1999; Toth et al., 2006). Over the long term, CPP has been shown to promote more positive peer relations in middle childhood in children of mothers with depression through its positive effect on attachment security in early childhood (Guild, Toth, Handley, Rogosch, & Cicchetti, 2017). These results suggest that the benefits of CPP are evident as many as 6 years post-intervention.

These studies are unique because there is limited research on the use of dyadic therapies specifically for mothers with depression and their children (Tsivos, Calam, Sanders, & Wittkowski, 2015). Rather, a significant proportion of the literature focuses on interventions that target maternal depressive symptoms in isolation (Poobalan et al., 2007). Although evidence-based individual psychotherapies have been shown to be effective in reducing mothers’ postnatal depressive symptoms (Fitelson, Kim, Baker, & Leight, 2011), research suggests that treatment of maternal depression alone may not be sufficient for preventing negative socioemotional development in their children (Forman et al., 2007). Relationally based treatments that focus on the mother-child dyad have been shown to be superior to individual psychotherapies for improving child outcomes and enhancing the quality of the mother-child relationship (Muzik et al., 2009). These findings suggest that CPP may be a particularly relevant intervention for improving the developmental outcomes of children of mothers with depression.

Although one previous study has examined the long-term effects of CPP on peer relationships in children of mothers with depression (Guild et al., 2017), to our knowledge, no other studies have examined the efficacy of CPP on maternal and child outcomes beyond 12 months post-intervention. By investigating the long-term effects of CPP on child affect, we can start to better understand the effect of targeting attachment in order to support the psychological development of children of mothers with depression. Highlighting the long-term effects of CPP on maternal affective expression and dyadic cohesion also provides important information about the sustained behavioral impact of dyadic interventions in the parenting context, particularly among mothers with depression.

Hypotheses

This study will examine the longitudinal effects of CPP on the following constructs: a) maternal affective expression, b) child affective expression, and c) cohesion within the mother-child dyad. Specifically, it will examine whether change to secure attachment post-intervention mediates the association between baseline group and long-term outcomes. A previously published study in the current sample showed that children of mothers who participated in CPP showed greater change to secure attachment post-intervention compared to children of mothers in the depressed and non-depressed control groups (Toth, Rogosch, Manly, & Cicchetti, 2006). The present study hypothesizes that this change to secure attachment following participation in CPP will be associated with higher levels of maternal warmth, child positive affect, and dyadic cohesion, and lower levels of maternal hostility and child negative affect/behaviors when children are 9 years old.

Method

Participants

Participants included a subset of mothers and children (N = 135) who took part in a follow-up assessment of a randomized controlled trial evaluating the efficacy of CPP for toddlers of mothers with depression (see Toth et al., 2006). Recruitment targeted mothers with a history of major depressive disorder (MDD) since giving birth to their child (N = 130). To minimize co-occurring risk factors that may accompany maternal depression (Coyne & Downey, 1991), families could not be reliant on public assistance and mothers were required to have at least a high school education. Mothers were recruited through referrals from mental health professionals and through notices placed in newspapers and community publications, in medical offices, and on community bulletin boards. Maternal psychopathology was assessed via the Diagnostic Interview Schedule, Version III, Revised (DIS-III-R; Robins, Helzer, Cottler, & Goldring, 1988). Mothers who met criteria for bipolar disorder were excluded from participation; however, those with other comorbid Axis I disorders (e.g., anxiety, substance use disorder, schizophrenia, etc.) were retained. Anxiety disorders (53.8%), bulimia (11.5%), and alcohol-related disorders (9.2%) were among the most prevalent comorbid diagnoses. Mothers were 92.8% White, 3.5% Black, 2.1% Hispanic, and 2.3% other, and ranged in age from 21 to 41 years (M = 31.68, SD = 4.68). Children were 47.2% female and 20.34 months old on average (SD = 4.68).

Mothers with depression and their toddlers were randomized to either a depressed intervention condition (DI, n = 66), in which dyads received CPP, or a depressed control condition that did not receive an intervention (DC, n = 64). T1 assessments revealed that comorbid Axis I disorders were common, with 59.1% of mothers in the DI condition and 71.9% of mothers in the DC condition having at least one comorbid disorder. Over the course of the study, mothers in the depressed conditions were not restricted from pursuing other mental health treatment for themselves and/or their toddlers if desired. Thus, during the period between T1 and T2 assessments, 46.7% (DI) and 44.2% (DC) of mothers received some form of outside intervention, including individual psychotherapy (34.0%), marital therapy (11.3%), group therapy (5.2%), and family therapy (2.1%). Mothers also were not restricted from taking psychotropic medication; however, receipt of medications was not routinely monitored. Group contrasts revealed no significant differences between the DI and DC groups on the receipt of other mental health treatment (Toth et al., 2006).

Mothers without depression or other history of major mental illness (NC, n = 68) and their toddlers were also recruited to serve as an additional comparison group. Names of potential families with a toddler age child were obtained from birth records. Mothers were recruited directly by contacting families of similar socioeconomic status who lived in the vicinity of the mothers with depression. Mothers in the NC group were screened with the DIS-III-R and excluded if they met criteria for the presence of current or past major psychiatric disorder. See Table 1 for a description of participant demographic variables across groups.

Table 1.

Group Comparison of Demographic Characteristics

	Baseline Group			Group Contrasts
	DC (n = 64)	DI (n = 66)	NC (n = 68)
Demographics	M (SD)	M (SD)	M (SD)	df	F	p
BDI score	16.91 (9.26)	15.44 (9.15)	2.60 (3.37)†	1,128	.83	.37
Age	30.66 (4.91)	31.55 (4.87)	32.78 (4.03)	2,195	3.54	.03
Annual family income	45,727 (11,402)	48,485 (11,224)	48,559 (10,462)	2,195	1.39	.25
Yrs of education	14.95 (1.79)	15.33 (1.56)	14.71 (1.70)	2,195	2.35	.10
	n (%)	n (%)	n (%)	df	χ2	p
Race/ethnicity				10	10.97	.36
White, non-Hispanic	56 (87.50)	63 (95.45)	65 (95.59)
White, Hispanic	1 (1.56)	0 (0)	0 (0)
Black, non-Hispanic	3 (4.69)	2 (3.03)	2 (2.94)
Black, Hispanic	2 (3.13)	1 (1.52)	0 (0)
Pacific Islander	0 (0)	0 (0)	1 (1.47)
Other	1 (3.13)	0 (0)	0 (0)

Note. DC = Depressed Control. DI = Depressed Intervention. NC = Non-Depressed Control.

^†BDI scores for mothers in the NC group were not included in the group contrast

Procedure

Prior to study participation, mothers provided informed consent for their and their children’s participation. Research was conducted with University of Rochester Institutional Review Board approval. Trained interviewers who were unaware of mothers’ depressive status conducted all assessments. Mothers were provided with a small financial incentive for their time.

Baseline (T1) assessments were conducted when toddlers were 20 months old on average (M_age = 20.34 months, SD = 2.50, N = 198). During an initial home-based session, mothers were administered a demographics interview and maternal mental health was assessed using the DIS-III-R and the Beck Depression Inventory (BDI; Beck, Ward, Mendelson, Mock, & Erbaugh, 1961). In a subsequent laboratory-based session, mothers and toddlers participated in the Strange Situation paradigm (Ainsworth, Blehar, Waters, & Wall, 1978) as a measure of toddler attachment security.

Follow-up assessments (T2) were conducted when toddlers reached approximately 36 months old (M_age = 38.23 months, SD = 2.85, N = 163) and participants in the CPP condition had completed the preventive intervention. Mothers were again administered the BDI and DIS-III-R. During a subsequent laboratory-based session, mother-toddler dyads completed the Strange Situation.

A final follow-up (T3) took place when children were on average 9 years old (M_age = 9.77 years, SD = 0.65, N = 135). At this time, mothers again provided consent for their and their child’s participation in the IRB-approved follow-up study. Children provided their assent. Families were given a small financial incentive for their participation. Mothers were administered the BDI-II (Beck et al, 1996). In a laboratory-based session, mothers and children were videotaped during a 6 minute long structured interaction task (Mother-Child Interaction; Granger, Weisz, McCracken, Ikeda, & Douglas, 1996).

Participant Retention

Between T1 and T2, 35 families were lost to follow-up. As a result of failing to complete all of the T2 assessments or moving out of area, 8 dyads were lost from the DI group, 8 from the DC group, and 4 from the NC group. Additionally, 12 mothers assigned to the DI condition either declined to participate in CPP or discontinued their participation early. Finally, the T2 Strange Situation recordings could not be coded for 2 of the DC dyads and 1 NC dyad due to technical difficulties. As a result, group sizes for the cases who completed T2 assessments were as follows: DI, N = 46; DC, N = 54; NC, N = 63. No significant differences due to completion status were found for baseline maternal depression scores, or for demographic characteristics, including maternal age, educational level, marital status, or annual family income (Toth et al., 2006). Additionally, no completion status effects were found with respect to child gender or baseline attachment classification, and mothers in the depressed conditions who were lost to follow-up did not differ significantly with respect to the recency of MDD, onset, severity, or comorbidity (Toth et al., 2006). Therefore, there was no evidence of selection bias in the retained sample of T2 completers.

Of the 163 mother-child dyads who completed T2 assessments, 28 did not participate in the T3 follow-up. Attrition was primarily due to families moving out of the area, and there were also some MCI videos that could not be coded due to technical difficulties. Group sizes for the 135 dyads who completed T3 were as follows: NC, N = 46; DC, N = 47; DI, N = 42. To assess for the possibility of differential attrition, group contrasts were performed via Student’s t-tests and chi-square testing. No significance differences between completers and non-completers were found for baseline maternal age (t(161) = 1.95, p = .16), race (χ²(4, N = 135) = 2.14, p = .71), level of education (t(161) = .20, p =.66), marital status (χ²(3, N = 135) = 3.46, p = .33), or annual family income (t(161) = .03, p =.87). Missing data was therefore determined to be missing completely at random.

Preventive Intervention: Child-Parent Psychotherapy

CPP is an attachment theory-informed preventive intervention for young children ages zero to five and their caregiver(s) (Lieberman, Ghosh Ippen, & Van Horn, 2015). The primary goals of the intervention are to 1) enhance the caregiver’s capacity to respond in developmentally appropriate ways to the child’s needs for nurturance, socialization, and protection; 2) modify distorted or maladaptive perceptions that the caregiver or child may have about each other; 3) restore the child’s sense of safety and trust in the caregiver as a secure base for meeting the child’s needs; and 4) promote the caregiver’s ability to support the child’s autonomy, while balancing both his/her own needs and those of the child. Sessions took place at the clinic in weekly hour-long sessions.

In the current investigation, CPP was initiated following randomization to the DI group and completion of baseline assessments. The intervention period averaged 58.19 weeks (SD = 10.00). Although sessions were scheduled weekly, the mean number of sessions conducted was 45.24 (SD = 11.16; range = 30 – 75), due to cancellations and missed appointments. To ensure fidelity of the intervention, implementation of CPP was monitored through weekly individual and group supervision, discussions of videotaped therapy sessions, and monthly monitoring of videotaped sessions for each case. The intervention was implemented by licensed masters level therapists.

Measures

Diagnostic Interview Schedule, Version III, Revised (DIS-III-R; Robins et al., 1988).

The DIS-III-R is a structured psychiatric interview designed to assess for Axis I disorders in accordance with the diagnostic criteria of the Diagnostic and Statistical Manual of Mental Disorders, Third Edition, Revised (DSM-III-R; American Psychiatric Association, 1987). The interview is organized into modules that assess for symptoms of DSM-III-R Axis I disorders. Items are rated on a yes/no basis and diagnoses are generated by computer algorithms, thereby precluding the need for interrater reliability. Previous test-retest reliability studies on the DIS-II for DSM-III diagnoses in current or former psychiatric populations yielded adequate kappas of .50 or greater (with the exception of panic disorder [.40]) (Robins, Helzer, Ratcliff, & Seyfried, 1982). The DIS-III-R was used to assess for major depressive disorder at baseline to determine diagnostic inclusion criteria in the current study.

Beck Depression Inventory (BDI and BDI-II; Beck et al., 1961, 1996).

The BDI is a widely used self-report measure that assesses for symptoms of depression. Its 21 items are rated on a 0 to 3 scale, with 3 indicating the most intense symptom severity over the past week. For the present study, the BDI was used to measure the severity of maternal depressive symptoms at baseline and T2 follow-up. The BDI has been shown to correlate with psychiatric ratings of depression in both psychiatric and student samples (Beck et al., 1988). Test-retest reliabilities range from .48 to .86 among psychiatric patients and .60 to .83 among non-psychiatric populations (Beck et al., 1988). The revised version of the BDI, the BDI-II, was used to assess maternal depression at T3. A meta-analysis of 118 studies that reported psychometric properties of the BDI-II indicated test-retest reliabilities ranging from .73 to .96 across psychiatric, medical, and non-clinical samples (Wang & Gorenstein, 2013). The BDI-II also demonstrates strong convergent validity with other widely used measures of depression and anxiety (Wang & Gorenstein, 2013). Internal consistencies for the BDI in the present sample were α = .88, α = .89, and α = .90 at T1, T2, and T3 respectively.

Strange Situation (Ainsworth et al., 1978).

The Strange Situation is a paradigm designed to observe attachment relationships between caregiver and child. Based on the child’s behavior in response to a series of separations and reunions with the mother and a stranger, the child can be classified into one of four attachment styles: insecure-avoidant (A), secure (B), insecure-ambivalent (C), or disorganized (D). For the purposes of the present study, attachment types A, C and D were collapsed into a single category representing insecure attachment.

Two trained raters who were unaware of diagnostic status and group assignment coded all videotaped sessions conducted at baseline. Agreement on classifications in the current sample was 90%. During the course of coding, raters discussed cases that were either considered to be unclear or complex (19% of sessions). The Ainsworth et al. (1978) criteria for attachment types A, B, and C were supplemented by a developmental systems approach for children ages 18 to 24 months (Gersten, Coster, Schneider-Rosen, Carlson, & Cicchetti, 1986). The modified framework accounts for the developmental reorganization that occurs within the attachment behavioral system, as well as between the attachment, exploration, affiliation, and fear/wariness behavioral systems, as children become older. Mean levels of proximity and contact seeking, contact maintenance, avoidance, and resistance are adjusted accordingly. Unlike younger children, children within this age range have developed language that can be additionally assessed in each of these categories. Type D classifications were determined using the Main and Solomon (1990) criteria.

Strange Situations conducted with 3-year-olds at T2 were coded using the MacArthur Preschool Attachment Classification System (Cassidy & Marvin, 1992). To serve as an additional check on reliability, a third rater who had no previous involvement with the investigation coded 20% of all videos. Coder agreement was 94% (K = .907).

Mother-Child Interaction (MCI; Granger, Weisz, McCracken, Ikeda, & Douglas, 1996).

For the present study, the MCI utilized a standard “revealed differences” paradigm (Strodtbeck, 1951), which requires two people to first make individual decisions based on their shared experiences, and then to attempt to reconcile their differing opinions. The interaction task was adapted from the work of Granger and colleagues (1996), who utilized a revealed differences paradigm with mothers and their children. This task has been successfully used to assess mother-child conflict with children as young as 5 years old (Granger, Serbin, Schwartzman, Lehouz, Cooperman, & Ikeda, 1998).

Before completing the MCI, mothers and their children completed two warm-up tasks to start a conversation with one another: planning a meal for a large family and planning a family vacation. These interactions were not coded, but rather were included to prevent the conflict task from occurring in isolation.

Mothers and children were then administered an “issues checklist” from which they were asked to independently rate 14 topics thought to be common sources of parent-child conflict (e.g., doing homework, bedtime, household chores, etc.). For each item, mother and child were asked to use a 5-point Likert scale (1 = never; 5 = constantly) to indicate “how much of a problem” the particular topic was at home. Research assistants then summed the mother and child ratings for each topic, and based on the highest score, selected the topic judged by the dyad to lead to the highest levels of conflict at home. If the summed scores for two topics were equivalent, the topic that presented first on the list was chosen for the interaction task. Mothers and children were then asked to sit across from each other and told by a research assistant that they would be given 6 minutes to discuss the chosen topic of conflict, with “the goal of coming to a solution.” All interactions were videotaped and were coded according to the system described next.

The System for Coding Interactions in Parent-Child Dyads (SCIPD; Lindahl & Malik, 1996) & The System for Coding Interactions and Family Functioning (SCIFF; Lindahl & Malik, 2001).

The MCI was coded using a combination of codes from the SCIPD and SCIFF, both of which are global behavioral coding systems. The SCIFF is a companion manual to the SCIPD and the majority of codes between the two systems overlap. The SCIPD was designed to behaviorally assess parent-child interactions during a challenging or frustrating experience. The following categories and codes were used in the current study: a) parental affect modulation (i.e., Negative Affect, Positive Affect, and Withdrawal), b) the quality of the parent’s responses towards the child (i.e., Emotional Support, Rejection/Invalidation, and Coercive Control), c) the child’s ability to modulate affect (i.e., Negative Affect, Positive Affect, Withdrawal, and Oppositionality/Defiance), and d) the quality of the interaction within the parent-child dyad (i.e., Cohesiveness, defined as the degree of unity, togetherness, and closeness between the parent and child). However, Maternal Withdrawal was excluded from analyses due to extremely low levels of withdrawal observed across mothers.

The Anger/Frustration and Sadness codes from the SCIFF were used to supplement the SCIPD to describe child negative affect. A preliminary examination of the data revealed that displays of sadness were quite rare, whereas anger was a much more prevalent emotion. Due to the low frequency of sadness, the decision was made to exclude the Sadness code from analyses. Additionally, given that the SCIFF defines Negative Affect as the combined display of sadness, anger, and anxiety, this code was also excluded from analyses.

Codes from the SCIPD and SCIFF are rated on a Likert scale from 1 (very low) to 5 (high), indicating the degree to which each observed behavior is characteristic of the parent, child, or of the dyad as a whole over the span of the entire interaction. In making ratings, coders must consider both the frequency and intensity of behaviors. The SCIFF has been demonstrated to have adequate interrater reliability, with Pearson correlation coefficients for individual codes ranging from .59 to .92 (Lindahl & Malik, 2001). All interactions were coded by the first author. For the purpose of attaining inter-rater reliability, 25 percent of interactions were also coded by a trained research assistant. Raters were unaware of group status and met for weekly meetings until adequate interrater reliability was achieved. Consensus ratings were utilized during final data analyses, but were excluded from calculations of interrater reliability. Kappas for the present sample ranged from K = .82 to K = .89.

Data Analytic Plan

Confirmatory factor analysis (CFA) was performed to evaluate the factor structure of the T3 maternal variables (N = 135). Five of the maternal emotions and behaviors coded during the T3 Mother-Child Interaction task (Positive Affect, Negative Affect, Emotional Support, Rejection/Invalidation, and Coercive Control) were examined. Zero-order correlations among the observed maternal affective, child affective, and dyadic study variables from the SCIPD and the SCIFF are presented in Table 2. MPlus Version 7.0 statistical software (Muthén & Muthén, 1998–2019) was used to test the factor structure of the variables. Missing data was handled using maximum likelihood estimation (ML), which uses each case’s available data to compute the value of the parameter that is most likely to have resulted in the observed data. Model fit was assessed using chi-square goodness of fit, the root mean square error of approximation (RMSEA), the comparative fit index (CFI), and the standardized root mean square residual (SRMR). A non-significant χ² statistic, RMSEA values less than or equal to .06, CFI values greater than .95, and SRMR values less than or equal to 0.07 were considered indicative of good model fit (Hu & Bentler, 1999; Yu and Muthén, 2002).

Table 2.

Correlations among T3 Study Variables

Variable	1	2	3	4	5	6	7	8	9	10	11
1. Maternal positive affect	--
2. Maternal emotional support	.76***	--
3. Maternal negative affect	−.61***	−.78***	--
4. Maternal rejection/invalidation	−.63***	−.76***	.83***	--
5. Maternal coercive control	−.51***	−.58***	.71***	.79***	--
6. Child positive affect	.74***	.70***	−.57***	−.59***	−.47***	--
7. Child anger	−.48***	−.57***	.62***	.62***	.53***	−.53***	--
8. Child oppositionality/defiance	−.43***	−.55***	.56***	.44***	.34***	−.44***	.55***	--
9. Child withdrawal	−.29***	−.44***	.48***	.37***	.32***	−.46***	.44***	.47***	--
10. Dyadic cohesion	.80***	.85***	−.82***	−.80***	−.69***	.77***	−.71***	−.63***	−.54***	--
11. Maternal depressive symptoms	−.20*	−.06	.16	.15	.27**	−.14	.19*	.01	.05	−.17	--

^*p < .05.

^**p < .01.

^***p < .001.

Structural equation modeling (SEM) was used to assess the main study hypotheses (N = 198; n = 35 missing on all variables except exogenous variables). Two dummy coded variables were created to represent T1 intervention condition: the first compared dyads in the DI group to those in the DC group (DI = 0, DC = 1), and the second compared dyads in the DI group to those in the NC group (DI = 0, NC = 1). The dummy variables representing the T1 intervention condition were modeled as exogenous predictors of T2 attachment insecurity (secure = 1, insecure = 0), T2 maternal depressive symptoms, and each of the five T3 outcome variables (Child Positive Affect (manifest variable), Child Anger/Problem Behaviors (latent factor), Maternal Warmth (latent factor), Maternal Hostility (latent factor), and Dyadic Cohesion (manifest variable)). T2 attachment security was modeled to predict each of the five T3 outcome variables. To control for baseline attachment, T1 attachment security was modeled to predict T2 attachment security. Receipt of other treatments during the intervention period (yes = 1, no = 0) was modeled to predict T2 depressive symptoms to control for the effects of other interventions, because mothers were not prohibited from pursuing additional interventions. T1 maternal depressive symptoms were also modeled as an exogenous predictor of T2 attachment security. In addition, maternal depressive symptoms at T1 were modeled to predict symptoms at T2, and symptoms at T2 modeled to predict symptoms at T3 to control for changes in depression over time. T2 maternal depressive symptoms were correlated with T2 attachment security, and also served as a predictor of the five T3 outcome variables. Finally, all T3 outcome variables were modeled to covary with each other. Given that T2 attachment security was represented categorically, pathways were estimated via the weighted least squares mean- and variance-adjusted approach (WLSMV) using Mplus, Version 7 statistical software (Muthén & Muthén, 1998–2019). The WLSMV is a robust estimator, which does not assume normally distributed variables and is ideal for modeling categorical data.

To determine whether change in T2 attachment mediated the relationship between T1 intervention condition and T3 outcomes, the distribution of the product of the coefficients method was employed with 95% asymmetric confidence intervals (RMediation; Tofighi & MacKinnon, 2011). Due to the fact that mediated effects are rarely normally distributed, 95% asymmetric confidence intervals were utilized to test the indirect effects (MacKinnon, 2008). Confidence intervals that do not include the value zero were considered statistically significant.

Results

Preliminary Analyses

To assess for group differences, analyses comparing mean differences on the main study variables were examined (see Table 3). At T1, 83.3% of dyads were insecurely attached in the DI group. After participating in CPP, 32.6% of DI dyads were insecurely attached. By contrast, few dyads changed from insecure to secure in the DC group, with 78.1% of dyads insecurely attached at T1 and 83.3% insecurely attached at T2. In the NC group, rates of insecure attachment were much lower and also showed stability, with 44.1% of dyads coded as insecure at T1 and 52.4% coded as insecure at T2. Mean levels of maternal depressive symptoms did not significantly differ between the DC and DI groups and were significantly higher than the NC group across time. There were no significant group differences on any of the T3 SCIFF/SCIPD outcome variables. Greater T3 maternal depressive symptoms were significantly correlated with less positive maternal affect, greater maternal coercive control, and greater child anger (see Table 2).

Table 3.

Group Comparisons among Main Study Variables

	DC	DI	NC
	M (SD)	M (SD)	M (SD)	Group Contrasts	p
Insecure T1	78.1%	83.3%	44.1%	X(2, N=198)=28.05	<.001
Secure T1	21.9%	16.7%	55.9%
Insecure T2	83.3%	32.6%	52.4%	X(2, N=163)=27.00	<.001
Secure T2	16.7%	67.4%	47.6%
Maternal Depressive Symptoms T1	17.39 (9.29)	16.20 (9.60)	2.43 (3.09)	F(2,1) = 68.75	<.001
Maternal Depressive Symptoms T2	9.94 (7.85)	10.39 (8.99)	2.24 (2.49)	F(2,1) = 26.62	<.001
Maternal Depressive Symptoms T3	12.96 (12.41)	11.48 (10.16)	2.72 (3.55)	F(2,1) = 17.81	<.001
Maternal Positive Affect T3	2.51 (0.95)	2.67 (1.12)	2.91 (1.07)	F(2,1) = 1.74	0.18
Maternal Emotional Support T3	2.72 (1.23)	2.79 (1.07)	2.87 (1.15)	F(2,1) = 0.19	0.83
Maternal Negative Affect T3	2.45 (1.32)	2.45 (1.25)	2.11 (0.95)	F(2,1) = 1.26	0.29
Maternal Rejection/Invalidation T3	2.34 (1.31)	2.24 (1.25)	2.11 (1.14)	F(2,1) = 0.41	0.66
Maternal Coercive Control T3	1.75 (1.21)	1.50 (0.86)	1.39 (0.75)	F(2,1) = 1.64	0.20
Child Positive Affect T3	2.55 (1.02)	2.62 (1.17)	2.67 (1.03)	F(2,1) = 0.10	0.90
Child Anger T3	2.09 (1.06)	2.14 (1.12)	2.04 (0.89)	F(2,1) = 0.21	0.81
Child Oppositionality/Defiance T3	1.98 (1.38)	1.83 (1.06)	1.98 (1.13)	F(2,1) = 0.88	0.42
Child Withdrawal T3	2.28 (1.35)	2.62 (1.21)	2.50 (1.15)	F(2,1) = 0.15	0.86
Dyadic Cohesion T3	3.47 (1.27)	3.38 (1.31)	3.67 (1.12)	F(2,1) = 0.67	0.52

Note. DC = Depressed Control. DI = Depressed Intervention. NC = Nondepressed Control.

Measurement Modeling

First, a CFA was tested with all maternal codes loading onto a single factor. This model showed poor fit to the data (χ²(4) = 29.851, p < .001; RMSEA = .219; CFI = .952; SRMR = .034). Next, a two-factor structure was tested such that Positive Affect and Emotional Support were indicators of the first latent factor, “Maternal Warmth,” and Negative Affect, Rejection/Invalidation, and Coercive Control served as indicators for the second latent factor, “Maternal Hostility.” The two latent factors were modeled to correlate with each other and residual covariance was modeled between Rejection/Invalidation and Coercive Control. This model showed good fit to the data (χ²(3) = 5.95, p = .11; RMSEA = .09; CFI = 1.00; SRMR = .01) and was a significantly better fit than the single factor model (Δχ²(1) = 23.9, p < .001). The indicator variables for Maternal Warmth and Maternal Hostility had strong factor loadings and the residual correlation between Rejection/Invalidation and Coercive Control was significant. The correlation between the two latent factors was also significant (see Figure 1 for coefficients).

Figure 1. CFA of Maternal SCIFF/SCIPD Variables.

Note. Pos Aff = Positive Affect. Emo Supp = Emotional Support. Neg Aff = Negative Affect. Rej/Inval = Rejection/Invalidation. Coerc Ctrl = Coercive Control.

***p < .001.

To examine the factor structure of the T3 child emotions and behaviors coded during the MCI, a second CFA was performed on child Positive Affect, Anger/Frustration, Oppositionality/Defiance, and Withdrawal. Child Positive Affect was represented as a manifest variable, and child Anger/Frustration, Oppositionality/Defiance, and Withdrawal were modeled as indicators of a latent factor, “Child Anger and Problem Behavior.” This latent factor was modeled to covary with Child Positive Affect. The model was a good fit to the data (χ²(2) = 2.53, p = .28; RMSEA = .04; CFI = 1.00, SRMR = .02). Factor loadings were strong and Child Anger and Problem Behaviors correlated significantly with Child Positive Affect (see Figure 2 for coefficients).

Figure 2. CFA of Child SCIFF/SCIPD Variables.

Note. Prob Beh = Problem Behavior. Opp/Def = Oppositional Defiance. Wthdr = Withdrawal.

***p < .001.

Structural Equation Modeling

SEM results indicated that the model was a good fit to the data (χ²(69) = 69.33, p = .47; RMSEA = .01; CFI = 1.00, WRMR = .45). There was evidence of significant stability of attachment security from T1 to T2. Consistent with previously published data using this sample (Toth et al., 2006), T1 intervention group uniquely predicted change to secure attachment at T2 after controlling for T1 attachment. Specifically, toddlers in the DI group were significantly more likely than those in the DC group to demonstrate change to secure attachment at T2. Furthermore, toddlers in the DI group were significantly more likely than those in the NC group to demonstrate change to secure attachment at T2. Both the NC and DC groups showed stability in attachment from T1 to T2. Change in T2 attachment significantly predicted Maternal Warmth at T3, such that mothers of toddlers who changed to secure attachment at post-intervention showed higher levels of warmth at T3. Change in T2 attachment also had a significant effect on Child Anger/Problem Behaviors at T3, such that children who changed to secure attachment at T2 exhibited less anger and fewer problem behaviors at T3 than those who did not change attachment classification. The effects of change in T2 attachment on T3 Maternal Hostility and Child Positive Affect were nonsignificant. There were no direct effects of T1 intervention condition on T2 maternal depressive symptoms or any of the five T3 outcomes. Maternal depressive symptoms were stable over time, and did not have a significant effect on change in T2 attachment security. Mothers who participated in other mental health treatments were more likely to have greater depressive symptoms at T2. The residual correlation between change in T2 attachment security and T2 maternal depressive symptoms was nonsignificant. Greater maternal depressive symptoms at T2 predicted significantly less Maternal Warmth at T3. T2 maternal depressive symptoms significantly predicted T3 Child Anger/Problem Behavior and T3 Dyadic Cohesion such that children of mothers with higher levels of depressive symptoms exhibited more anger and problem behavior and lower cohesiveness with their mothers, respectively. Finally, the residual correlations between outcome variables were all significant (see Table 4 for path coefficients and Figure 3 for significant paths).

Table 4.

Standardized Path Coefficients for Structural Model

Path			β	SE	p
DC vs. DI	→	Secure Attachment T2	−.57	.09	<.001
	→	Maternal Depressive Symptoms T2	−.05	.07	.45
	→	Maternal Warmth T3	.07	.13	.61
	→	Maternal Hostility T3	−.03	.12	.84
	→	Child Positive Affect T3	.00	.12	.99
	→	Child Anger/Problem Behavior T3	−.25	.14	.09
	→	Dyadic Cohesion T3	.12	.11	.29
NC vs. DI	→	Secure Attachment T2	−.34	.15	.03
	→	Maternal Depressive Symptoms T2	−.22	.15	.16
	→	Maternal Warmth T3	.08	.15	.60
	→	Maternal Hostility T3	−.06	.15	.68
	→	Child Positive Affect T3	−.13	.14	.38
	→	Child Anger/Problem Behavior T3	−.03	.16	.89
	→	Dyadic Cohesion T3	.10	.14	.48
Secure Attachment T1	→	Secure Attachment T2	.47	.09	<.001
Maternal Depressive Symptoms T1	→	Secure Attachment T2	.07	.16	.67
	→	Maternal Depressive Symptoms T2	.51	.07	<.001
	→	Receipt of Other Treatment	.19	.08	.02
Secure Attachment T2	→	Maternal Warmth T3	.30	.12	.02
	→	Maternal Hostility T3	−.18	.13	.15
	→	Child Positive Affect T3	.15	.12	.23
	→	Child Anger/Problem Behavior T3	−.39	.14	.01
	→	Dyadic Cohesion T3	.22	.12	.07
Maternal Depressive Symptoms T2	→	Maternal Warmth T3	−.19	.09	.05
	→	Maternal Hostility T3	.10	.08	.23
	→	Child Positive Affect T3	−.12	.09	.20
	→	Child Anger/Problem Behavior T3	.20	.09	.03
	→	Dyadic Cohesion T3	−.18	.08	.02
	→	Maternal Depressive Symptoms T3	.58	.06	<.001
Maternal Depressive Symptoms T2	←→	Secure Attachment T2	.14	.11	.20
Maternal Depressive Symptoms T3	←→	Maternal Warmth T3	−.04	.08	.62
	←→	Maternal Hostility T3	.13	.08	.08
	←→	Child Positive Affect T3	−.05	.07	.51
	←→	Child Anger/Problem Behavior T3	.06	.09	.53
	←→	Dyadic Cohesion T3	−.07	.07	.27
Maternal Warmth T3	←→	Maternal Hostility T3	−.84	.04	<.001
	←→	Child Positive Affect T3	.81	.04	<.001
	←→	Child Anger/Problem Behavior T3	−.73	.07	<.001
	←→	Dyadic Cohesion T3	.94	.02	<.001
Maternal Hostility T3	←→	Child Positive Affect T3	−.60	.06	<.001
	←→	Child Anger/Problem Behavior T3	.79	.05	<.001
	←→	Dyadic Cohesion T3	−.89	.02	<.001
Child Positive Affect T3	←→	Child Anger/Problem Behavior T3	−.67	.08	<.001
	←→	Dyadic Cohesion T3	.76	.04	<.001
Child Anger/Problem Behavior T3	←→	Dyadic Cohesion T3	−.90	.04	<.001

Note. DC = Depressed Control. DI = Depressed Intervention. NC = Non-Depressed Control.

Figure 3. Path Model Depicting Standardized Estimates of Significant Pathways.

Note. DI = Depressed Intervention. DC = Depressed Control. NC = Non-Depressed Control. Prob Beh = Problem Behavior. Standard errors are in parentheses. Contemporaneous residual correlations are not shown for the sake of simplicity.

*p < .05. **p < .01. ***p < .001.

Mediation Analyses

Results from the mediation analyses indicated that when comparing mother-child dyads in the DI and DC conditions, change to secure attachment at T2 significantly mediated the effect of baseline group assignment on T3 Maternal Warmth (95% CI [−.63, −.04]). Specifically, compared to toddlers of mothers with depression who did not receive any intervention, insecurely attached toddlers of mothers with depression who received CPP were more likely to change to secure attachment at T2, and in turn, their mothers showed greater levels of warmth at T3. Change to secure attachment at T2 also significantly mediated the effect of baseline group assignment on T3 Child Anger/Problem Behavior (95% CI [.08, .76]), such that compared to toddlers of mothers with depression who did not receive any intervention, insecurely attached toddlers of mothers with depression who received CPP had higher rates of change to secure attachment at T2, and in turn, exhibited lower levels of anger and problem behavior at T3.

When comparing mother-child dyads in the DI group to those in the NC group, change to secure attachment at T2 marginally significantly mediated the effects of baseline group assignment on T3 Maternal Warmth (96% CI [−.43, −.003]). In addition, change to secure attachment at T2 did significantly mediate the relationship between baseline group assignment and T3 Child Anger/Problem Behavior, such that compared to children of mothers without depression, children of mothers with depression who received CPP were more likely to change to secure attachment at T2, and in turn, exhibit fewer anger and problem behaviors at T3 (95% CI [.004, .54]). Finally, change in T2 attachment did not mediate the association between baseline group assignment and T3 Dyadic Cohesion (95% CI = [−.51, .02]).

Discussion

The overarching goal of this research was to better understand the efficacy of CPP for improving maternal and child affect and interactional patterns over an extended period of child development, particularly among mothers with depression and their children. To do this, change in attachment security at post-intervention was evaluated as a mediator of the relationship between intervention condition and maternal and child affective and relational outcomes in middle childhood.

As shown in previous studies with the present sample (Toth et al., 2006), children of mothers with depression who received CPP attained significantly higher rates of secure attachment post-intervention than children of mothers with depression who received no intervention and children of mothers without depression. The present study added to previous findings by showing that participation in CPP predicts maternal warmth and child anger/problem behavior when children are 9 years old through its effect on attachment security at 36 months. These results indicate that CPP continues to have positive indirect effects on both caregivers and children as far as six or more years post-intervention.

The long-term effects of CPP on maternal warmth have important implications for both maternal behavior and children’s developmental outcomes. The longitudinal findings from the present study suggest that participation in CPP may have fostered positive maternal behavior (i.e., maternal warmth) that improved attachment security, and this change in maternal behavior is stable over time. Increases in maternal warmth have significant implications for children’s continued development as it is associated with a number of positive child outcomes, including decreased responsiveness to fear and stress (Kuhlman, Olson, & Lopez-Duran, 2013), better social competence (Altschul, Lee, & Gershoff, 2016), fewer behavior problems (Eiden, Edwards, & Leonard, 2007), and increased trust in parents and peers in adolescence (Umemura & Serek, 2016).

Findings suggest that CPP also demonstrated long-term effects on child anger/problem behavior at 9 years old. These findings are consistent with previous research demonstrating that children with an early secure attachment to their caregiver(s) exhibit fewer externalizing and internalizing behaviors later in childhood (Fearon et al., 2010; Groh et al., 2012). Child anger and behavior problems at age 9 were also directly predicted by maternal depressive symptoms post-intervention, which is consistent with previous research showing the effects of maternal depression on the development of child externalizing behavior (Barker, Copeland, Maughan, Jaffee, & Uher, 2012; Goodman et al., 2011). These findings emphasize the need for early intervention for maternal depression in order to prevent negative child developmental outcomes, perhaps in addition to dyadic mother-child interventions.

However, findings from the present study did not support the hypothesis that CPP’s effect on attachment would contribute to the reduction of maternal hostility or increase in positive child affect when children were 9 years old. Because the Mother-Child Interaction task was designed to provoke conflict within the dyad, it is possible that in this context, mothers who may have typically displayed lower overall levels of hostile behavior became more difficult to distinguish from those with chronically higher levels of hostility due to the level of conflict elicited by the paradigm. Additionally, mothers were aware of being filmed during the interaction, which raises the possibility that they may have tried to regulate their displays of hostility in an effort to present themselves in a favorable light. On the whole, children also seemed to exhibit more negative or neutral affect during the MCI, which was likely a function of the fact that many of them explicitly stated at the outset of the task that they did not want to talk about the selected topic of conflict with their mothers. It is therefore possible that the MCI task was not ideal for eliciting maternal hostility or positive affect in children.

Although findings showed that dyadic cohesion when children were 9 years old was not predicted by attachment security at 36 months, dyadic cohesion was predicted by maternal depressive symptoms post-intervention. These results suggest that maternal depression in early childhood may more strongly impact cohesive mother-child interactions than attachment. Past literature has also shown the impact of maternal depression on parenting behavior (Goodman & Gotlib, 2002), although the effect of depression on mother-child cohesion in middle childhood has not been explicitly studied.

Limitations

Although the present investigation has a number of strengths, there are also some limitations that are important to acknowledge. First, the sample was restricted to middle-class mothers in an effort to minimize confounding risk factors. Additionally, the majority of mothers were White and had graduated from college. Therefore, our findings may not generalize to children and mothers of different racial and ethnic backgrounds, or to those from lower socioeconomic strata. However, previous research has shown CPP to be efficacious in fostering secure attachment and reducing behavior problems in samples of diverse racial and socioeconomic backgrounds (Cicchetti et al., 2006; Lieberman, Ippen, & Van Horn, 2006). Thus, these findings are promising with respect to CPP’s potential to promote adaptive developmental trajectories in more diverse populations.

An additional limitation relates to mothers being free to seek other services for their depression if desired. As it would have been unethical to prevent mothers with depression from obtaining treatment, it is possible that additional services sought and/or use of psychotropic medication may have contributed to children’s improved social functioning. To control for this confound, we included use of other treatments as a predictor of T2 depressive symptoms in the structural model. In addition, contrasts revealed no significant differences between the DI and DC groups with respect to services sought.

Further, the absence of data on intervening events post-intervention presents the possibility that alternative unmeasured variables, and not treatment effects, may explain maternal and child outcomes at T3. For example, it is possible that ongoing maternal depressive symptoms, changes in the family, changes in social support, or unmeasured differences in mothers who completed the intervention could better predict maternal and child affect.

Finally, in this investigation CPP was compared to a non-intervention control condition rather than another active treatment. As a result, it is difficult to determine whether findings are unique to an attachment theory-informed intervention. Despite this limitation, prior investigations have shown CPP to be more effective than psychoeducational parenting interventions in sustaining effects on secure attachment and fostering positive representations of the self and caregiver in children (Cicchetti et al., 2006; Stronach et al., 2013; Toth et al., 2002).

Conclusions

The findings of the present investigation support the efficacy of CPP as an intervention that fosters adaptive developmental trajectories in mothers and their children through the promotion of a secure attachment relationship in early childhood. Thus far, CPP is the only relational intervention for young children and their caregivers for which a follow-up of this length has been conducted. As an attachment-based intervention, CPP is therefore compellingly effective in promoting adaptive developmental trajectories and buffering against the sequelae of maternal depression. This study has shown that through the promotion of attachment security, CPP for toddlers and their mothers with depression leads to lasting benefits for children as well as their caregivers, providing at-risk children with the chance to develop greater social and emotional competence.