Coparental Discord and Children’s Behavior Problems: Children’s Negative Family Representations as an Explanatory Mechanism

Abstract

This study examined children’s insecure representations of the family as a mechanism accounting for the association between coparental discord and children’s externalizing problems in a diverse sample of two hundred and forty-three preschool children (M age = 4.60 years). Using a multi-method, multi-informant, prospective design, results indicated that coparental discord was indirectly related to children’s externalizing behaviors through their insecure representations of the family. Higher levels of coparental discord were specifically linked with more insecure representations of the family, which in turn, predicted higher levels of externalizing behaviors two years later. These pathways remained robust even after considering the roles of general family adversity, child gender, and family income per capita as predictors in the analyses.

Coparenting relationship quality, defined as the ability of caregivers to coordinate with one another and support each other in their roles and responsibilities as parents, is posited to have significant consequences for children’s adjustment. Due to the challenging nature of raising young children together, breakdowns in the coparenting relationship are relatively common and typically manifest in poor cooperation, increased conflict, and high levels of disengagement between parents over child-rearing issues. In highlighting its developmental significance, coparenting discord has been consistently documented as a risk factor for children’s psychopathology, particularly in the form of externalizing symptoms (McHale & Rasmussen, 1998; Schoppe, Mangelsdorf & Frosch, 2001). However, although coparenting discord has been a consistent and unique predictor of child externalizing behaviors, little is known about why this risk is pronounced. To address this gap in the literature, the current study examined children’s internal representations or appraisals of the family unit as a potential mediating mechanism underlying the association between coparenting discord and children’s externalizing problems.

According to family systems theory, the coparenting subsystem is relatively distinct from the interparental and parent-child relationships in its composition (Feinberg, 2003; Minuchin, 1985). As a basis for comparison, the interparental subsystem is comprised of only the adult partners in the family, operates at the dyadic two-person level, and includes the intimate relationship properties of the adult partners (Minuchin, 1985). Individual parent-child relationships are comprised of one parent and the child, operate at the dyadic level, and include only interactions between one parent and the child. By way of differentiation, the coparenting system is comprised of both parents and the child, operates at the whole-family level, and includes parent-parent interactions about child-rearing issues, parent-child interactions about the other parent, and most importantly, parent-parent-child interactions (Cabrera, Scott, Fagan, Steward-Streng, & Chien, 2012). In addition to its documentation as a distinct subsystem within the larger family system, coparenting quality has been shown to be a unique predictor of children’s later behavior problems. Meta-analytic findings have indicated that coparenting discord is a significant risk factor for children’s behavior problems during the preschool to early school years (Teubert & Pinquart, 2010). Conflict and diminished coordination between coparents in their shared role as caregivers have been concurrently and prospectively linked to externalizing behaviors in children during early and middle childhood (Jones, Shaffer, Forehand, Brody, & Armistead, 2003; Lewin, Mitchell, Beers, Feinberg, & Minkovitz, 2012; McHale & Rasmussen, 1998; Parkes, Green, & Mitchell, 2018; Schoppe et al., 2001). In addition, researchers have found that coparenting quality still accounted for unique variance in children’s later externalizing behaviors even after controlling for certain aspects of parental characteristics (e.g., well-being), caregiving practices (e.g., rejection), and interparental relationship quality (Dopkins-Stright & Neitzel, 2003; McHale & Rasmussen, 1998).

Repeated documentation of coparenting discord as a distinct predictor of child behavior problems raises the question of how and why the stressfulness of witnessing difficulties between parents in their caregiving roles may pose a direct risk to children. However, little is known about the specific child psychological mechanisms that may mediate the vulnerability experienced by children exposed to coparental discord. In the only study to test child characteristics as explanatory mechanisms underpinning coparenting quality, Brody and Flor (1996) reported that youth self-regulation partially accounted for the concurrent link between coparenting difficulties (i.e., lack of communication, diminished support, and increased conflict about child-rearing issues) and their impairments in psychosocial and academic competence. However, the relatively modest size of the indirect effects in their study suggests that there may be other important child characteristics underlying the risk associated with coparenting discord. Therefore, expanding the search for the child characteristics that account for the distinct association between coparenting difficulties and children’s behavior problems remains a critical research direction.

Although there is little in the way of formal, cohesive theory on why child mechanisms may mediate the risk posed by coparenting difficulties, one common explanation in the literature is that children’s vulnerability to coparenting discord may stem from their insecurity in the family system (Brody & Flor, 1996; Cummings, Koss & Davies, 2015; Forman & Davies, 2005). For example, in addressing why self-regulation accounted for the vulnerability experienced by youth exposed to coparenting difficulties, Brody and Flor (1996) proposed that children’s insecurity in the family was a key operative mechanism. In elaborating on this premise, emotional security theory (EST) posits that family discord undermines children’s natural tendencies to draw on the family unit as a source of security (Davies & Cummings, 1994). According to EST, children’s insecurity in the family system is commonly defined as recurrent doubts about the abilities of their parents to serve as consistent sources of stability, safety, and support (Davies & Cummings, 1994). By extension, EST proposes that children’s difficulties obtaining a sense of security in the family mediates the link between family problems and children’s later mental health (Davies & Cummings, 1994; Davies, Martin, & Sturge-Apple, 2016). Therefore, in the first link of the mediational chain, family conflict is proposed to serve as a source of interpersonal threat that undermines children’s security in the family. Children’s insecurity, in turn, is proposed to increase their vulnerability to externalizing problems.

A key indicator of children’s sense of security in the family is the quality of their internal representations of family relationship dynamics (Sturge-Apple, Davies, Winter, Cummings, & Schermerhorn, 2008). Negative internal representations of the family are often characterized by children’s negative appraisals of the meaning family difficulties have for the welfare of themselves and their families. These negative appraisals commonly consist of child portrayals of aggression between parents, harsh and unpredictable behavior from parents, and parental difficulties coordinating support and resources in a way that reliably provides safety and security for the child (Sturge-Apple et al., 2008; Yoo, Popp, & Robinson, 2014). Internal representations are proposed to develop from the children’s prior family experiences and serve as future guides for processing the meaning and consequences of interpersonal interactions and behaviors in negative ways that increase their behavioral problems (Sturge-Apple et al., 2008).

As a critical dimension of family discord, we hypothesized that coparental discord increases children’s risk for behavior problems by increasing their tendencies to develop negative, insecure representations of the family. Given the triadic nature of the coparenting relationship, children are directly immersed in the dynamics between their parents and, as a result, highly sensitive to problems in the relationship (Grych & Fincham, 1990). For example, children experience more distress, perceived threat, and self-blame when they are exposed to adult disagreements about child-rearing in comparison to disagreements involving interparental relationship issues (Grych & Fincham, 1993). Coparenting difficulties may also signal to children that parents are unable to coordinate their support in a way that reliably fulfills the child’s needs for security and predictability in the family unit. The resulting negative internal representations of the family may, in turn, increase children’s risk for behavior problems (Davies et al., 2016). For example, several prevailing theories (i.e. EST, risky family process) propose that children’s insecure representations of the family have the potential to generalize into broader adjustment problems if children use these internal models to guide their interpersonal interactions outside the family (Davies & Cummings, 1994; Repetti, Taylor, & Seeman, 2002).

Building on theoretical bases for insecure internal representations as a risk mechanism underpinning coparental discord, our study was designed to provide a first test of the explanatory role of children’s negative representations in the association between coparental discord and child behavior problems. We tested our hypothesis during the transitional period between the preschool and early school years for several reasons. First, the coparenting relationship is proposed to be especially salient during this period as young children spend most of their waking time with their parents (Cabrera et al., 2012; Feinberg, 2003). Second, the period of early childhood is posited to be a switch point for the consolidation of strategies for coping with future challenges based on the child’s prior experiences with support and adversity in the family (Del Giudice, Angeleri, & Manera, 2009). As part of this process, the development and internalization of stable, coherent internal representations of the family and the self in the family is hypothesized to be a key stage-salient process during this period. For example, increases in social perspective-taking abilities during early childhood are posited to sensitize children to concerns about the consequences family conflict have for the welfare of themselves and the family and, as a result, their safety and security (Davies et al., 2016). Third, novel interpersonal challenges (e.g., peers, structured classroom settings) during the transition to school are proposed to increase the likelihood of children relying on their previously formed representations of family relationships as a guide to processing and responding to novel social situations (Davies et al., 2016). Finally, research has shown that externalizing problems in childhood are consistently associated with poor outcomes in adolescence and adulthood, including conduct problems, antisocial behavior, substance use, and emotional difficulties (Reef, Diamantopoulou, van Meurs, Verhulst, & van der Ende, 2011; Silver, Measelle, Armstrong, & Essex, 2005).

In summary, although coparenting has been a consistent, robust predictor of child externalizing problems, little is known about how and why coparental discord is a distinct precursor of children’s behavior problems. We specifically focused on children’s externalizing symptoms based on several considerations. Coparenting discord has been identified as a particularly robust predictor of children’s externalizing symptoms relative to other forms of maladjustment (McHale & Rasmussen, 1998; Schoppe et al., 2001). For example, meta-analytic findings revealed that the magnitude of the effect sizes for associations between coparenting discord and children’s social, emotional, and behavioral problems were particularly high for children’s externalizing symptoms (Teubert & Pinquart, 2010). Moreover, even internalizing difficulties experienced by young children are commonly manifested in externalizing symptoms characterized by heightened irritability and anger (e.g., Luby, Heffelfinger, Mrakotsky, Hessler, Brown, & Hildebrand, 2002). In further underscoring its developmental significance, research has shown that externalizing symptoms during early childhood are robust precursors of numerous psychological difficulties and behavioral problems in adolescence and adulthood (Odgers, Moffitt, Broadbent, Dickson, Hancox, Harrington…& Caspi, 2008; Reef et al., 2011; Silver et al., 2005).

To test children’s insecure representations of the family as an explanatory mechanism in the association between coparental discord and later child behavior problems, the current study employed a multi-method (observation, interview, survey), multi-informant (experimenter, parent, teacher, child), longitudinal design with two measurement occasions spaced two years apart. We specifically examined whether children’s insecure family representations assessed in a story-stem task at Wave 1 accounted for the prospective association between observations of coparental discord in a triadic interaction and teacher reports of child behavior problems two years later. The use of teacher reports was designed to increase rigor in our hypothesis testing by examining whether the role of negative internal representations as a risk factor expands beyond the family to predict children’s difficulties in school settings. To increase the rigor of the analyses, we examined whether the indirect path involving coparental discord, children’s internal representations, and their externalizing problems was significant even after controlling for earlier behavior problems and children’s exposure to adversity in other family subsystems. Given that interparental relationship quality, parenting practices, and parent psychopathology have been identified as predictors of children’s insecure representations and their behavior problems (Breaux, Harvey & Lugo-Candelas, 2016; Davies et al., 2016; Sturge-Apple et al., 2008), we utilized a relatively comprehensive adversity composite of these family dimensions as a covariate in the analyses. Finally, child gender and family income per capita were also included as covariates in the model based on prior research linking them with coparenting quality or child externalizing behaviors (Davies & Lindsay, 2001; Dodge, Pettit, & Bates, 1994; Kolak & Vernon-Feagans, 2008; McHale & Rasmussen, 1998).

Methods

Participants

Data for this study was drawn from a larger, multi-method, multi-informant study designed to examine the impact of the quality of interparental relationships on children’s coping and mental health. This study included data from two measurement occasions spaced two years apart, beginning when children were in preschool. Participants were recruited through universal pre-K programs, preschools, community centers, and child care centers in a Northeastern metropolitan area. Inclusion criteria for the study were as follows: (1) adult caregivers were raising a child together as romantic partners and had frequent contact with each other as a triad (i.e. 2-3 times per week for at least one year), (2) both adult caregivers and the child were willing to participate, and (3) the child was within one year of enrolling in kindergarten. At Wave 1, participants included 243 families (mothers, partners, and children). The average age of children at Wave 1 was 4.6 years old (SD = 0.44 years), with 56% of the sample consisting of females. Almost half the sample was Black or African American (48%), with smaller percentages identifying as White (43%), multiracial (6%), or another race (3%). The median household income was $36,000 per year (range = $2,000 – $121,000) with most families (69%) receiving public assistance. The median education level of parents in the study was a high school diploma or GED. At Wave 1, 93% of parents were living together (average length = 3.36 years). Most mothers (99%) and their partners (74%) were biological parents of the child. At Wave 1, 47% of the parents in the sample were married, and 99% of the parents were in heterosexual relationships (3 lesbian couples). The retention rate from Wave 1 to Wave 2 was 91%.

Procedure and Measures

At each of two measurement occasions, parents and children participated in two visits to a research laboratory with visits spaced about one week apart. The laboratory included two rooms: (a) an observation room designed to resemble a family room and equipped with audiovisual equipment to capture and record family interactions, and (b) an interview room where parents completed confidential interviews and questionnaires. All procedures were approved by the IRB prior to conducting the current study. Parents were compensated monetarily for their time and participation, and children received small prizes at each visit.

Coparenting Quality.

At Wave 1, both parents and the child participated in a Family Interaction Task (FIT) designed to capture coparenting interactions. The family was instructed to build a LEGO house together during the 10-minute task using a picture of a prototype house to guide them. The task was designed to elicit individual differences in coparenting behavior as children were unable to successfully build the Lego house without parental structuring and assistance. This family interaction task was video recorded and later coded by trained coders using an adaptation of the Coparenting and Family Ratings System (CFRS; McHale, Kuersten-Hogan, & Lauretti, 1999). Given that the CFRS was originally designed to assess coparenting and family functioning among married couples in middle and upper-middle class families, we made three primary modifications to the CFRS: (1) excluded codes (e.g., parent-child warmth) that did not specifically capture coparenting characteristics; (2) developed a coparenting disengagement code based on recent refinements in observational assessments of coparenting (McHale, Salman-Engin, & Coovert, 2015); and (3) expanded the rating systems and descriptive anchors from the 5- to 7-point to 9-point scales to more sensitively capture differences in coparental functioning in our more demographically diverse sample. Prior research has shown the CFRS to have strong construct and predictive validity (McHale, Kuersten-Hogan, & Lauretti, 2001).

Trained coders rated coparenting conflict, disengagement, warmth, and cooperation along a nine-point dimensional rating scale (1 = not at all characteristic; 9 = highly characteristic). Coparenting conflict captures the amount of disagreement, insult, and sarcasm between parents, with high ratings defined by multiple instances of unambiguous criticism between parents that are hostile in nature and disruptive to family functioning. Coparenting disengagement is characterized by parental apathy, unresponsiveness, and detachment from each other during the task with the child, with higher ratings reflecting frequent instances of these behaviors from one or both parents throughout the session in a way that disrupts the family interaction. Coparenting warmth assesses the level of positive regard between parents in their interactions with their child, with higher scores reflecting strong cohesion and teamwork in coordinating their roles as parents and engaging their child in the task. Finally, coparenting cooperation captures active, overt cooperation between parents during play with their child, with higher scores reflecting frequent support of the other parent and constructive engagement between partners in their child-rearing activities. Two trained coders independently coded all 243 family visits. As indices of interrater reliability, average-measure intraclass correlation coefficients (ICC) using a two-way random model with absolute agreement ranged from .73 to .81 for the four coparenting codes. The four coparenting codes were used as indicators of the latent construct, coparenting discord. For the purposes of the current study, both coparenting warmth and cooperation were reverse-scored so that higher scores reflected lower levels of warmth and cooperation between parents.

Internal Representations.

At Waves 1 and 2, children participated in a revised version of the MacArthur Story Stem Battery (MSSB-R; Cummings, Schermerhorn, Keller, & Davies, 2008), a narrative storytelling technique designed to capture children’s internal representations of family relationships. Narrative story-telling techniques, such as the MSSB-R, have been regarded as more comprehensive measures of child representations than other measures of child assessment (i.e.: survey, semi-structured interview) as they are able to tap into both implicit and explicit appraisals (Fiese & Spagnola, 2005; Robinson, 2007). In addition, these procedures capitalize on children’s natural interests and skills in story-telling, lending validity to representation assessments (Bretherton, Ridgeway, & Cassidy, 1990). In support of its validity, the MSSB-R codes have been consistent correlates of child adjustment problems (Bascoe, Davies, Sturge-Apple, & Cummings, 2009). Following MSSB-R procedures, an experimenter presented the child with six story stems depicting various stressors and conflicts in interparental, mother-child, and partner-child relationships. To foster child engagement, experimenters used animated, dramatic voices, various toy props, and family action figures that resembled the child’s sex and ethnicity. After experimenters administered each story stem, children completed the stories using the props and action figures to verbally and physically convey what happened next in the story between their family members. If needed, experimenters used standard probes to clarify child actions or prompt children to address the main issue in the story.

Video records of each story were later rated by trained coders along four seven-point dimensional rating scales of family cohesion, family antagonism, family apathy, and overall child insecurity (MSSB-R). First, family cohesion is characterized by children’s appraisals of the long-term positive impact of family challenges and stressors. High family cohesion ratings specifically reflect portrayals of close, supportive, and harmonious family relationships in the wake of the stressful events depicted in the story stem. For the current study, this scale was reverse-scored so that higher ratings reflected child narratives with few to no signs of positivity, closeness, or warmth between members of the relationship dyad. Second, family antagonism reflects the level of hostility, anger, and antipathy present in the relationship dyad, with high scores reflecting relationships that are predominantly hostile and combative. Third, family apathy gauges the degree of indifference, unresponsiveness, and detachment in the child’s portrayal of the relationship dyad, with higher scores reflecting relationships that are predominantly indifferent, cold, somber, or neglectful. Finally, overall child insecurity reflects the degree to which the portrayals of family events are likely to serve as a source of support or threat in the children’s goal of preserving their physical and psychological well-being. Evaluations of overall child insecurity are based on the overall quality and pattern of the child’s responses throughout each story stem. Ratings on each scale were aggregated across the six stories (M α = .74, range = .62 - .82), with the resulting four composites used as indicators of a latent construct of children’s insecure internal representations of the family. Because ratings were based on a single coder with two coders overlapping on 20% of the videos, interrater reliability indices were calculated based on single-rater ICCs using two-way random models with absolute agreement (Syed & Nelson, 2015). The resulting ICCs ranged from .69 to .92 across the four codes and two measurement occasions.

Externalizing Behaviors.

At Waves 1 and 2, children’s teachers completed the MacArthur Health and Behavior Questionnaire (HBQ; Ablow, Measelle, Kraemer, Harrington, Luby, Smider,…Kupfer, 1999). Child externalizing behaviors were derived from three HBQ subscales: oppositional defiance (9 items; e.g.: “This child has temper tantrums or a hot temper.”); overt hostility (4 items; e.g.: “This child taunts and teases other children.”); and conduct problems (6 items; e.g.: “This child destroys his or her own things.”). Teachers responded to each item using a 3-point scale (0 = never or not true, 2 = often or very true). Previous research supports the test-retest reliability and validity of the HBQ in clinical and community samples (Ablow et al., 1999; Essex, Boyce, Goldstein, Armstrong, Kramer, Kupfer, & MacArthur Assessment Battery Working Group, 2002). Internal consistencies for the scales used in this study ranged from .80 to .92. Ratings on the three subscales were used as indicators of latent constructs of child externalizing behaviors at each wave.

Family Adversity Covariate:

At Wave 1, we obtained maternal and partner reports of three dimensions of family adversity. First, to assess interparental relationship quality, couples completed the Couple Satisfaction Index (CSI; Funk & Rogge, 2007). The CSI is comprised of the mean of 4 items (e.g.: “I have a warm and comfortable relationship with my partner.”) that are each rated on 7-point scales (“not at all true” to “completely true”). Scales were reverse-scored so that higher scores indicated greater relationship dissatisfaction. The CSI scales have been shown to have convergent and discriminant validity and greater measurement precision in detecting differences in relationship satisfaction levels than other measures in the literature. Internal consistencies for maternal and partner reports were excellent (α = .99).

Second, mothers and partners completed the Socialization of Moral Affect Questionnaire (SOMA; Rosenberg, Tangney, Denham, Leonard, & Widmaier, 1994). The SOMA was designed to assess parental endorsement of specific child-rearing strategies to scenarios depicting child successes, failures, and transgressions (e.g., child loses hat and cannot recall what happened to it). Prior research has found the SOMA to be both a reliable and valid measure of parenting behaviors (Rosenberg, 1998; Rosenberg et al., 1994). Parents rated the likelihood they would respond in various ways to each scenario along a five-point scale (1 = not at all likely, 5 = likely). Five subscales were selected to assess parenting difficulties: neglect/ignoring (8 items, e.g.: not making a big deal about your child’s great performance in his or her lead role in the school play), conditional approval (8 items; e.g.: "You know how important it is that you do well in school to make me happy."), power assertion (8 items; e.g.: "You are going to sit here at this table until you finish this meal, and I don't care if you are here all night long!”), disgust/teasing (7 items, e.g.: giving your child a disgusted look and saying, “That makes me sick.”), and love withdrawal (8 items, e.g.: “You are not acting like someone I want to play with,” and leaving the game.). Responses across the five subscales were averaged together to obtain maternal and partner reports of their parenting difficulties. Internal consistencies for the maternal and partner reports on the overall parenting difficulties scales were .73 to .79, respectively.

Third, to assess parental psychopathology symptoms, mothers and their partners completed the Anxiety (12 items; “I am so tense in certain situations that I have great difficulty getting by”), Depression (12 items; “I’ve forgotten what it’s like to feel happy.”), and Antisocial Behavior (12 items; “I get a kick out of doing dangerous things.”) scales from the 63-item Personality Assessment Inventory (PAI; Morey, 1991). Items on the scales are rated on a four-point scale, ranging from 1 (“False, not at all true") to 4 (“Very true"). Research supports the reliability and validity of the PAI scales in both community and clinical samples (Deisinger, 1995; Morey, 1991). To maximize the comprehensiveness and parsimony of the parent psychopathology assessments, responses to the items across the three subscales were averaged together to create indicators of maternal and partner psychopathology symptoms. Internal consistencies for the maternal and partner psychopathology measures were .73 to .70, respectively.

Mother’s and partner’s scores on each of the six indicators (mother and partner ratings of interparental relationship satisfaction [reversed scored], parenting difficulties, parent psychopathology) were standardized and then averaged together to create a parsimonious composite of family adversity for the primary analyses (α = .63).

Covariates: Demographic characteristics.

Measures of family income per capita (total annual family income/number of people living in the household) and child sex (1 = female, 2 = male) were also included as covariates in the model.

Analytic Plan

Variables were examined for normality, and correlations among indicators of latent constructs were examined for strength and direction. Structural equation modeling (SEM) within the Amos 25.0 statistical software program (Arbuckle, 2017) was used to examine the indirect path involving coparenting discord, children’s insecure representations of the family and their subsequent externalizing problems in the context of the family and child covariates. Model fit was assessed using the following fit statistics: χ², χ²/df ratio, comparative fit index (CFI) and root mean square error of approximation (RMSEA). CFI values ≥ .95 and RMSEA values ≤ .05 are seen as evidence for a well-fitting model, and CFI values > .90 and RMSEA values between .05 and .08 seen as evidence for adequately fitting models (Hu & Bentler, 1999). In addition, χ² values that are significant, as well as χ²/df ratio values < 3 are seen as evidence for well-fitting models (Tay & Drasgow, 2012).

To test our hypothesized process model, we specified Wave 1 coparenting discord as a predictor of children’s Wave 1 insecure representations of the family and their Wave 2 behavior problems. We also estimated a path between children’s Wave 1 insecure family representations and their Wave 2 externalizing behaviors. In addition, the three Wave 1 covariates (i.e., family adversity, income per capita, and child sex) were specified as predictors of children’s insecure family representations at Wave 1, and their behavior problems at Wave 2. Finally, we specified correlations between all Wave 1 exogenous variables and between error terms of the same manifest indicators of externalizing behaviors across the two time points (see Figure 1).

Figure 1:

A cross-lagged, autoregressive structural equation model examining the explanatory role of children’s insecure representations of the family in the association between coparenting discord and children’s externalizing behaviors. For clarity, significant pathways have been bolded and non-significant lines have been dashed. Parameter estimates for structural paths are standardized coefficients. Correlations were estimated among all predictors; however, only significant correlations are depicted. *p<.05.

Results

Descriptive Analyses

A summary of the means, standard deviations, and correlations for the primary variables in the study can be found in Supplementary Table 1. All variables in the primary analyses were normally distributed. Correlations among manifest indicators of latent constructs were moderate to strong in magnitude and in the expected directions. Wave 1 coparenting measures were significantly linked with indicators of children’s insecure internal representations and externalizing symptoms at Waves 1 and 2. Wave 1 insecure representations were also correlated with children’s externalizing symptoms in the expected directions. Finally, the Wave 1 family adversity and income covariates were significantly related to Wave 1 coparenting characteristics and some indices of children’s internal representations at Wave 1 and their externalizing problems at Wave 2. The percentage of missing data in our sample was 9%, ranging from 0% to 28% (i.e., teacher reports of externalizing symptoms at Wave 2). Missing data analyses examining the association between amount of missing data for each family and the variables in the primary analyses revealed no significant findings from the 21 analyses conducted. Given that maximum likelihood approaches for estimating data provide accurate parameter estimates for all types of missing data when the amount of missing data in the sample is below 20% (Schlomer, Bauman, & Card, 2010), we used full information maximum likelihood estimation (FIML) to estimate missing data and retain the full sample in our analyses.

Primary Analyses

The resulting model depicted in Figure 1 provided a satisfactory representation of the data, χ² (99, N = 243) = 203.49, p < .001; χ²/df ratio = 2.06, CFI = .94, and RMSEA = .07, using fit statistics outlined by Hu & Bentler (1999). The standardized loadings of manifest indicators onto their respective latent constructs of coparenting discord, children’s insecure representations of the family, and their externalizing behaviors were all significant (p < .001), in the expected directions, and moderate to strong in magnitude (M = .87). All estimated paths are shown in the figure, and for clarity, significant paths are bolded. Although the family adversity composite did not predict children’s insecure representations or behavior problems as a covariate, lower family income per capita was significantly related to children’s greater insecure representations of the family, ß = −.29, p < .001. Likewise, the significant association between child sex and insecure representations, ß = .18, p < .001, reflected that boys exhibited more negative family representations than girls. Of direct relevance to our hypotheses, Wave 1 coparenting discord was associated with higher levels of children’s insecure representations of the family at Wave 1, ß = .18, p = .01. Children’s negative representations, in turn, predicted increases in their externalizing behaviors at Wave 2, ß = .17 p = .04, even after controlling for the significant autoregressive path for children’s behavior problems, ß = .37, p < .001. Further supporting children’s insecure representations as an explanatory mechanism, the indirect path involving Wave 1 coparenting discord, Wave 1 insecure representations of the family, and their Wave 2 externalizing problems was significantly different from zero, with the unstandardized coefficient for the indirect path = .01, and the 95% CI [.0002, .0165].

Follow-Up Analyses

Empirical support for insecure family representations as an explanatory mechanism would be further bolstered by demonstrating that coparenting difficulties are prospectively related to children’s representations of family relationships. Therefore, as a further follow up analysis, we examined whether Wave 1 coparenting difficulties predicted children’s insecure family representations at Wave 2 while also specifying the three covariates in the primary analyses and the autoregressive path involving Wave 1 child negative representations as simultaneous predictors of Wave 2 child representations. Consistent with the primary analyses in Figure 1, we also specified correlations between all Wave 1 exogenous variables and between error terms of the same manifest indicators of insecure representations across the two time points. The resulting model depicted in Figure 2 provided a satisfactory fit with the data, χ² (74, N = 243) = 196.25, p < .001; χ/df ratio = 2.65, CFI = .92, and RMSEA = .08, according to fit statistics outlined by Hu & Bentler (1999). The standardized loadings of the manifest indicators onto their respective latent constructs of Wave 1 coparenting discord, and Waves 1 and 2 insecure representations were all significant (p<.001), in the expected directions, and moderate to strong in magnitude (M = .77). All estimated paths are shown in the figure and, for clarity, significant paths have been bolded. In support of its role as an intermediary process, the results showed that Wave 1 coparenting discord significantly predicted higher levels of children’s insecure representations of the family at Wave 2, ß = .16, p < .05.

Figure 2:

A follow-up cross-lagged, autoregressive structural equation model examining the explanatory role of children’s insecure representations of the family in the association between coparenting discord and children’s externalizing behaviors. For clarity, significant pathways have been bolded and non-significant lines have been dashed. Parameter estimates for structural paths are standardized coefficients. Correlations were estimated among all predictors; however, only significant correlations are depicted. *p<.05.

Discussion

Although coparenting conflict has been a consistent and unique predictor of children’s behavior problems (Teubert & Pinquart, 2010), little is known about how or why this association persists. Guided by emotional security theory (Davies & Cummings, 1994), this multi-method, multi-informant, longitudinal study examined children’s insecure representations of the family as an explanatory mechanism in the prospective association between coparental discord and children’s later behavior problems. Consistent with hypotheses, the SEM findings indicated that Wave 1 coparenting discord was concurrently associated with children’s insecure representations of the family. Children’s insecure representations of the family in turn predicted their higher levels of externalizing symptoms over a two-year period. The indirect path involving children’s insecure representations remained significant even after controlling for overall family adversity, demographic characteristics, and prior child behavior problems. In addition, follow-up findings indicating that coparenting discord predicted subsequent change in children’s insecure representations of the family two years later further supported the hypothesized temporal ordering of coparenting discord, children’s insecure family representations, and their behavior problems. In building on previous studies of coparenting, these findings provide a first empirical test of children’s representations of the family as a mechanism in the longitudinal association between coparental discord and children’s adjustment difficulties.

In addressing the first link in the proposed indirect pathway, our findings beg the question of why coparental discord may evidence unique associations with insecure family representations even after considering other family risk factors. According to Grych and Fincham (1990), conflicts about children may be more threatening or aversive to them than conflict that does not concern them (i.e. marital conflict). Therefore, coparenting difficulties may be more closely linked with children’s negative appraisals and coping difficulties than general marital conflict or dissatisfaction because children are more firmly entrenched as participants in the coparenting relationship than they are in the interparental relationship. Consistent with this premise, previous findings have indicated that parental conflicts about children elicit more fear, self-blame, and shame than conflicts about interparental issues (Grych & Fincham, 1993). Research has also shown that more intense child-centered conflict leads to higher levels of negative affect and perceived threat in children, suggesting that child-related coparenting conflict may be especially detrimental to children’s adjustment (Grych & Fincham, 1993).

As another plausible explanation for the unique association between coparenting discord and children’s insecure family representations, it is possible that coparenting difficulties signify to children that parents are unable to coordinate their caregiving practices in a way that reliably provides the support and structure necessary to promote children’s sense of safety and stability in the family unit. Coparenting discord may specifically reflect widespread parental difficulties encapsulating their conflicts within the interparental relationship. According to family systems theory, family interactions are governed by boundaries or implicit rules and guidelines for interacting within and across family subsystems (Minuchin, 1985). When discord is not successfully encapsulated within the specific subsystem in which it originated, it may contribute to disengaged or enmeshed family relationship structures. For example, the diffuse boundaries in enmeshed families are manifested in emotional entanglement between members across family subsystems (Coe, Davies, & Sturge-Apple, 2018). Proliferation of discord across family relationships is posited to carry more significant risks for children than discord that is confined to the interparental subsystem (Davies, Cummings, & Winter, 2004; Grych & Fincham, 1990).

Highlighting the second link in the indirect pathway of coparenting discord, our findings indicated that children’s insecure internal representations of the family significantly predicted higher levels of children’s externalizing behaviors two years later. As children transition into the early school years, they increasingly encounter novel and challenging interpersonal situations outside of the family (e.g., unfamiliar peers, formalized classroom settings, extracurricular activities). In these novel social situations, children may rely on their previously formed internal representations of the family to process and respond to novel interpersonal situations outside the family unit (Sturge-Apple et al., 2008). In reflecting the storage of relevant knowledge of interpersonal relationships, internal representations are posited to provide scripts for efficiently processing complex interpersonal dynamics and generating rapid interpretations and response strategies in social interactions (Dykas & Cassidy, 2011). Thus, children’s negative or insecure representations of themselves in the family may specifically guide their future interactions outside the family and increase their negative information processing and responses to these novel social situations (Bascoe et al., 2009; Dykas & Cassidy, 2011). Consistent with this explanation, prior research has shown that children with more negative family representations tend to process challenging extrafamilial events in more negative ways that increase their risk for subsequent psychological problems (Bascoe et al., 2009).

Although the current study provided support for the role of children’s insecure family representations as an intermediary mechanism underlying links between coparenting difficulties and children’s externalizing problems, several limitations warrant discussion in interpreting the results. First, in contrast to our empirical documentation of coparenting difficulties as a risk factor, the SEM results revealed negligible associations among the family adversity composite and children’s representations and later behavior problems. At some level, these results may be interpreted as inconsistent with previous findings indicating that interparental relationship satisfaction, parenting difficulties, and parent psychopathology are significant predictors of children’s feelings of security in the family and their externalizing behaviors (Davies et al., 2016; Sturge-Apple et al., 2008). However, at the correlational level of analysis (see Supplementary Table 1), family adversity was associated with children’s negative representations and externalizing problems. Thus, one plausible explanation for the inconsistency between our findings and the previous studies is that these family factors largely pose a risk through their shared variance with coparenting discord. Alternatively, although our measurement battery was designed to provide psychometrically sound measures of the variables in the primary analyses, it is also possible that the reliance on surveys to assess family adversity may have limited its power to predict child functioning. Likewise, although our goal was to provide a relatively comprehensive assessment of family adversity, future research would benefit from expanding measures of family covariates to include other possible confounding variables such as interparental conflicts about issues unrelated to parenting issues.

Second, although the current study provided a longitudinal test of the association between coparenting discord and child adjustment, we cannot conclusively test mediation without a fully-lagged model and a measure of the hypothesized mediating mechanism at a time point between Waves 1 and 2. Maxwell and Cole (2007) argue that without a fully lagged, longitudinal model, mediational analyses are likely to produce biased parameter estimates due to their lack of statistical control for the mediator at an earlier time point. Although our longitudinal analysis provided a first step towards the goal of testing mediation, our concurrent assessment of coparenting discord and children’s insecure representations of the family may have contributed to some biases in the estimation of pathways. Without the use of a fully-lagged mediational approach, a plausible alternative explanation is some of the documented associations in our study reflect bidirectional relationships between coparenting discord and children’s psychological functioning. However, through follow-up analyses, we were able to document a prospective link between Wave 1 coparenting discord and children’s Wave 2 insecure representations of the family. Thus, this increases the plausibility that a fully-lagged analysis may yield significant results supporting mediation. In the future, researchers should aim to examine whether coparenting discord predicts successive changes in both internal representations and children’s behavioral problems.

Third, although our selection of insecure representations of the family as an explanatory mechanism was specifically guided by EST, our findings do not diminish the salience of alternative risk processes posited in other theories. For example, according to social learning theory, witnessing family difficulties may increase children’s risk for behavior problems through vicarious learning processes. Thus, children may be more prone to emulate the antagonistic, unsupportive parenting behaviors displayed in coparenting interactions through the acquisition of abstract rules, scripts, and expectancies for enacting hostile behaviors (Bandura, 1983; Margolin, Oliver, & Medina, 2001). Therefore, it is possible that our assessment of insecure representations may be markers for the operation of these social learning processes. At a broader level, future process-oriented research would benefit from expanding the search for explanatory mechanisms and child sequelae. For example, given the modest magnitude of the parameter estimates in our model, it is possible that other potential mechanisms (e.g., child coping efficacy, parent emotion socialization practices, sibling relationship quality, parenting difficulties) may account for the longitudinal association between coparenting discord and children’s behavior problems. Likewise, expanding the scope of child functioning dimensions beyond our focus on externalizing symptoms (e.g., internalizing symptoms, social skills, academic competence) may offer important insights into the specificity and generalizability of our findings.

Despite these limitations, the current study provided a first test of children’s representations of family relationships as a mechanism underpinning the association between coparenting discord and children’s externalizing problems by following a diverse sample of families within a multi-method, multi-informant longitudinal design. Consistent with EST, the findings supported the notion that children’s insecure representations of the family operated as an explanatory mechanism in the association between family conflict and child adjustment (Davies & Cummings, 1994; 1998). As a further testament to the importance of the coparenting system in the development of children’s representations of the family, coparenting discord remained a significant predictor of children’s representations and behavior problems even after controlling for other aspects of family functioning. More broadly, the results highlight the value of assessing different aspects of family-level functioning (i.e. coparenting) and exploring potential child processes that may mediate the association between coparenting quality and child adjustment. From a translational perspective, our findings also underscore the potential utility of utilizing intervention programs designed to strengthen coparenting relationships (i.e.: Family Foundations – Feinberg, 2008; Building Strong Families – US Dept. of Health and Human Services, 2002). For example, these programs may ultimately reduce children’s psychological burdens and difficulties by improving parental abilities to constructively and cooperatively engage in coparenting interactions (Feinberg & Kan, 2008; Wood, McConnell, Moore, Clarkwest, & Hsueh, 2010). If future research provides further support for insecure representations as a risk mechanism underpinning coparenting discord, the findings may offer important guides to expanding clinical targets and tools for interrupting pathogenic processes associated with coparenting discord. For example, treatment programs for children experiencing difficulties between their parents (e.g., divorce, conflict) use structured play (e.g., puppets) activities and games as a way for children to safely express emotions and process and understand adverse family events. Adapting some of these components to coparenting interventions may facilitate the development of more balanced child representations of the family and themselves and provide a foundation for the use of more effective coping skills to reduce their psychological difficulties (Johnston, Roseby, & Kuehnle, 2009).