Longitudinal Effects of Post-divorce Interparental Conflict on Children’s Mental Health Problems through Fear of Abandonment: Does Parenting Quality Play a Buffering Role?

Abstract

In a sample of 559 children (ages 9–18), researchers investigated whether: (1) fear of abandonment mediated the association between post-divorce interparental conflict (IPC) and mental health problems, and (2) parent-child relationship quality moderated the association between IPC and fear of abandonment. Mediation analyses indicated that pretest IPC predicted fear of abandonment 3-months later, which then predicted child- and teacher-reported mental health problems 10-months later. The hypothesized protective effect of a high-quality parent-child relationship was not observed. IPC predicted fear of abandonment for all children, except for those with low- and moderate-quality father-child relationships, for whom IPC was not significantly related to fear of abandonment. Findings highlight the need to optimize child coping programs and improve parenting-after-divorce programs to reduce IPC.

Exposure to interparental conflict (IPC) is considered one of the most stressful events experienced by children whose parents divorce (Wolchik, Sandler, Braver, & Fogas, 1986). It is also the most well-documented predictor of children’s adjustment problems after divorce (Amato, 1993; Harold & Sellers, 2018; Johnston, Roseby, & Kuehnle, 2009). Decades of cross-sectional and longitudinal evidence demonstrate that exposure to IPC is detrimental for children’s functioning in multiple domains. Post-divorce IPC confers elevated risk for the development of mental and physical health problems (Amato & Cheadle, 2008; Fabricius & Luecken, 2007; Harold & Sellers, 2018; O’Hara, Sandler, Wolchik, & Tein, 2019), risky health behaviors (Elam et al., 2016; Orgilés, Carratalá, & Espada, 2015), and multiple aspects of developmental competence such as peer relationships (Forehand, Neighbors, Devine, & Armistead, 1994; Vandewater & Lansford, 1998), self-esteem (Noller, Feeney, Sheehan, Darlington, & Rogers, 2008), adaptive coping (Bing, Nelson, & Wesolowski, 2009), and academic achievement (Long, Slater, Forehand, & Fauber, 1988).

Over the past several decades, researchers have developed process-oriented models to elucidate mechanisms that may explain variability in the effect of IPC on children’s adjustment problems (Cummings & Davies, 2010, 2002; Harold & Sellers, 2018). One of the most well-studied models, Emotional Security Theory (EST), posits that exposure to IPC leads to children’s impaired sense of emotional security in the family (Cummings & Davies, 2010). Children who are emotionally insecure display intense and prolonged behavioral and affective signs of distress in response to IPC, which in turn predict later adjustment problems. The Cognitive Contextual Framework (CCF), a distinct but complementary model, also highlights the critical role of children’s appraisals of IPC (Grych & Fincham, 1990).

The context of parental divorce requires nuanced examination of the EST and CCF models to understand children’s responses to IPC. Prior studies of the EST and CCF models have been conducted with samples primarily or exclusively comprised of children in married or cohabitating families (for a notable exception see Cummings et al., 2011). Studying how parental divorce or separation shapes how children process and adjust to IPC is important given that children’s sense of security in the family and their appraisals of IPC are likely to be contextualized by their parents’ marital status. In married families, IPC is likely to lead to fears about parents getting divorced. For children from divorced families, IPC may create worries that they will lose contact with their nonresidential parent (Kurdek & Berg, 1987) or concerns that the remaining parent may also leave (Gardner, 1976). IPC may also cause a “pervasive sense of vulnerability as [children] feel that the protective and nurturant function of the family has given way,” inciting fear that their needs will be overlooked as their parents tend to their own distress (Wallerstein, 1985, p. 844). It is also possible that children who experience significantly less IPC after the divorce may actually feel a greater sense of security if the decrease in conflict allows the parents to focus more on the child’s needs.

For children from divorced families, a common threat appraisal that is directly related to their sense of security is the fear of losing contact or being abandoned by one or both parents (Gardner, 1976; Wallerstein, 1985). Fear of abandonment may be particularly salient when children are exposed to high levels of post-divorce IPC. IPC can put children in the middle of loyalty conflicts, disrupt the security of their continuing relationship with one or both parents, and undermine parental responsiveness (Buchanan, Maccoby, & Dornbusch, 1991; Krishnakumar & Buehler, 2000), all of which are likely to generate fear related to abandonment. Particularly in the context of divorce, parents often triangulate their children by denigrating the other parent in an attempt to create an alliance or undermine their relationship with the other parent (Rowen & Emery, 2019). Fear of abandonment has been shown to predict children’s mental health problems after divorce in several studies (Kurdek & Berg, 1987; Wolchik et al., 1993; Wolchik, Tein, Sandler, & Doyle, 2002). However, to our knowledge, no studies have tested the mediating role of fear of abandonment in the link between children’s exposure to post-divorce IPC and the development of mental health problems.

Process-oriented models also highlight the key role of moderators that attenuate the negative impact of IPC (e.g., Cummings & Davies, 2010). According to the EST, a child’s emotional security is influenced, in part, by the quality of the parent-child relationship (Davies & Cummings, 1994). Similarly, the CCF posits that contextual factors, such as children’s relationships with their parents, are critically important in shaping their appraisals of IPC (Grych & Fincham, 1990). High quality parenting is a robust protective resource for children, especially those who experience parental divorce (Sandler et al., 2012). How might high-quality parenting mitigate the effects of IPC on children’s fear of abandonment, and in turn, mental health problems? A strong sense of warmth and caring, a central feature of a high-quality relationship, is likely to attenuate the level of perceived threat during and after exposure to IPC by increasing children’s sense that they will be taken care of. Further, when parents are involved in their children’s daily lives, children are likely to believe that their parents will be available for them in a reliable manner, which may reduce fear of abandonment. Finally, children who feel close to their parent, accepted, and able to communicate openly are more likely to use their parent as a resource when distressed, which may reduce fear of abandonment in the context of IPC.

There are a few studies that have examined the capacity of high-quality parenting to buffer children from IPC in divorced families. In one study, researchers found that IPC and mother-child relationship quality independently predicted children’s adjustment after divorce in a counteractive manner (i.e., IPC predicted lower child adjustment whereas high relationship quality predicted higher child adjustment), but they did not interact (Lutzke, Wolchik, & Braver, 1996). Similarly, Sandler and colleagues (2013) found no evidence that IPC interacted with either mothers’ or fathers’ parenting quality to predict child mental health problems in a sample of high-conflict divorcing families. However, they did find a compensatory buffering effect of parenting in high-conflict families where high warmth of one parent protected against child internalizing problems but only if warmth of the other parent was low (Sandler, Miles, Cookston, & Braver, 2008). In addition, Brody and Forehand (1990) found that adolescents in families with high post-divorce conflict had fewer internalizing problems if they also had a good relationship with their non-residential father, as compared to those with a poor father-child relationship. Several researchers have investigated the capacity of high-quality parenting to buffer the effects of IPC in samples of married families. Some findings support the buffering hypothesis by showing that secure parent-child relationships reduced the strength of the relation between IPC and adjustment problems (El-Sheikh & Elmore–Staton, 2004; Hetherington et al., 1982; Katz & Gottman, 1997). Conversely, other studies have documented that although IPC and parenting both had main effects on mental health problems, parenting did not moderate the negative impact of IPC on mental health problems (Davies, Harold, et al., 2002; Grych, Raynor, & Fosco, 2004). In addition, a handful of studies indicate that the protective effect of parenting depends on factors, such as the child’s gender (Formoso, Gonzales, & Aiken, 2000) and type of outcome assessed (Crockenberg & Langrock, 2001).

Studies of children in married or cohabitating families that have examined the cognitive processes mediating the effect of IPC on children’s mental health problems have provided some evidence that high-quality parent-child relationships are protective. For example, DeBoard-Lucas and colleagues (2010) found that supportive parenting buffered the link between exposure to IPC and children’s perceptions of self-blame, but not their threat appraisals. Another study found that supportive parent–child relationships reduced both threat appraisals and self-blame related to IPC (Grych et al., 2004). In a sample of divorced families, Wolchik and colleagues (2002), found that the interaction between a composite of divorce stressors (which included IPC) and mother-child relationship quality on fear of abandonment was marginally significant; the relation between divorce-related stressors and fear of abandonment was attenuated at high levels of mother-child relationship quality.

Current Study

In the current study, we tested whether fear of abandonment mediated the association between exposure to post-divorce IPC and later mental health problems. We also examined whether a high-quality parent-child relationship buffered the effect of IPC on fear of abandonment, and whether mother-child relationships and father-child relationships provide different buffering effects. We assessed these relations longitudinally in a large, diverse sample of children ages 9–18 and used child-, parent-, and teacher-report of children’s mental health problems. Figure 1 illustrates the theoretical model and hypothesized effects. In our preregistered analytic plan (available at https://osf.io/sy93f/), we predicted that IPC would be positively associated with fear of abandonment, which would in turn be positively associated with children’s internalizing and externalizing problems. That is, we expected a significant indirect effect of IPC on children’s internalizing and externalizing problems through fear of abandonment. We further hypothesized that parent-child relationship quality would moderate the relation between IPC and fear of abandonment, such that the positive association between IPC and fear of abandonment would be attenuated at higher levels of parent-child relationship quality. We did not expect this protective effect of relationship quality to vary across mother- and father-child relationships. As per our preregistered plan, we tested two demographic variables, child age and ethnicity, as moderators of the relation between IPC and fear of abandonment. Neither age nor ethnicity were significant moderators (See results in Supplemental Material).

Figure 1.

Hypothesized theoretical model among study variables. A = A path; B = B path; Z = moderation path; (+) = hypothesized positive association; (−) = hypothesized negative association

Method

Participants

Participants were 559 children and their parents from a larger randomized controlled trial of the New Beginnings Program, a parenting-after-divorce program, conducted between 2012 – 2015 across four counties (two urban and two rural) in the Southwestern United States. The original sample included 830 families with 1,529 children ages 3–18; the current analyses included the 559 children ages 9–18 (nested in 353 families) who completed at least one interview. This subsample represents 73.8% of the 757 children who were eligible to be interviewed (ages 9–18 years) and whose parents consented for their child to be interviewed. Parents of children who participated were more likely to be older (M = 40.24 vs. 34.88 years, t = 9.99, p < .001), have higher pre-divorce income (M = $70k -$80k vs. $60k - $70k, t = −3.99, p < .001) and have more children (M = 2.18 vs. 1.53, t = −10.92, p < .001) than those whose children were eligible but did not participate in the interview. Of the 559 children, 201 had their father as the program parent and 358 had their mother as the program parent; the program parent provided data on child mental health problems. Note that in 40 families, both mothers and fathers participated in the trial; data from the first parent who enrolled in the trial were used. Teacher data were collected only with parents’ permission. Of the 559 children, 425 (76%) had parental permission for collecting teacher report and 375 (67.1% total; 88% of those with permission) had teacher report of their mental health problems.

Children were 12.48 years old (SD=2.60) on average and 50.6% were male. Approximately 93% of the participating parents had been legally married to their child’s other parent and 41.7% were legally divorced at study onset. On average, children spent 2.44 (SD = 2.64; range = 0–7) nights per week, or approximately 35% of overnights, with their fathers. The racial and ethnic composition of the parents was: 61.7% non-Hispanic White, 29.5% Hispanic, 2.9% Black, 1.8% Asian, and <1% American Indian or Native Alaskan; 3.9% identified as multiracial.

Procedure

The primary recruitment method was a 12-minute informational video, which was shown during a brief program that was mandated for all parents seeking a divorce or separation. Attempts were made to contact interested parents (N = 4,227) by phone. Of these parents, 2,155 were reached and screened for eligibility. Eligibility criteria were: (1) filing for divorce or modification of a divorce decree within the past two years or, if never married, changing a parenting time agreement within the past two years after separation; (2) at least one child between the ages of 3–18; (3) parent spends three or more hours per week or one overnight or more every other week with the child; (4) parent and child could complete interviews and questionnaires in English; and (5) family was not referred by the juvenile court or child protection services. The assessments were administered at pretest, posttest, and 10-month follow-up.

The sample consisted of 830 families that were randomized to the New Beginnings Program (NBP; 10-session skills-focused parenting program; n = 445) or the low dose control condition (LDC, 2-session psychoeducational parenting program; n = 385) in which either the mother (n = 474; 57.1%) or father (n = 356; 42.9%) enrolled. See Sandler et al (2020) for more information about recruitment and enrollment for the trial. In NBP, 80% (n = 358) completed the posttest, and 68% (n=304) completed the 10-month follow-up. In LDC, 83% (n = 320) completed the posttest and 78% (n = 299) completed the 10-month follow-up. Parents were asked for permission to interview children between ages 9–18. Of the 757 age-eligible children, 559 (73.8%) received permission and completed at least one interview; 515 (92%) completed the pretest, 437 (78%) completed the posttest, and 407 (73%) completed the 10-month follow-up. We conducted attrition analyses to determine if there were any differences in parenting, IPC, intervention condition, and demographic indices for those who participated at the 10-month follow-up. The only difference that was significant was that parents in the NBP condition were more likely to participate in the 10-month follow-up, F (1, 505) = 5.74, p = .017 than those in the LDC condition.

Interviews took place over the phone. Mothers, fathers, and children were interviewed by trained program staff masked to condition. Teachers completed online or mailed questionnaires. Parents provided informed consent and children provided informed assent. Parents were paid $50 and children were paid $30 for interviews. Teachers were paid $5 in cohort 1 and $10 in cohorts 2–4. All study procedures were approved by the university’s Institutional Review Board.

Measures

Interparental Conflict (IPC; child-report).

At pretest, children’s perception of IPC in the past month was assessed by 13 items (α=.87) from the frequency and intensity subscales of the Child Perception of Interparental Conflict scale (CPIC; Grych et al., 1992). Examples of frequency and intensity items are: “You never saw your parents arguing or disagreeing,” and “Your parents got really mad when they argued.” The CPIC is well-established in its reliability and validity (Bickham & Fiese, 1997; Grych et al., 1992). Children also completed the seven-item Caught in the Middle scale (CIM; Buchanan et al., 1991, α=.77). A sample item is: “How often does your mother/father ask you to carry messages?” This scale has been found to have good reliability and validity in samples of youth who have experienced parental divorce (Buchanan et al., 1991). Further, children completed two items related to badmouthing of the other parent (e.g., “My mother said bad things about my father”). In previous measurement work (see Sandler et al., 2020), a second-order one-factor model (using ordinal indicators and WLSMV estimation) using the above measures was conducted to test the dimensionality of IPC. The model fit the data well ($\mathrm{{\rm X}}$²[247] = 670.832, RMSEA = .058, CFI = .943, WRMR = 1.376) and factor scores were derived. These scores were used in the current study.

Parent-child Relationship Quality (child-report).

Pretest parent-child relationship quality over the past month was assessed by child report of three subscales from the Children’s Report of Parenting Behaviors Inventory (Schaefer, 1965): acceptance (16-item, α=.95; “Your mother/father made you feel better after talking over your worries with her/him”), rejection (16-item, α=.88; Your mother/father often acts as though you are in the way”), and consistent discipline (8-item, α=.78; “When your mother/father made a rule for you, she/he made sure it was followed”). The items were rated on a 5-point Likert scale from 1 (almost never) to 5 (almost always). Adequate reliability and validity are established for the CRPBI; scores have been associated with children’s adjustment problems in samples of divorced families (Wolchik, Wilcox, Tein, & Sandler, 2000). Children completed the 10-item Open Communication subscale of the Parent-Adolescent Communication Scale (PACS; Barnes & Olson, 1982, α=.92). Items (e.g., “Your mother/father tried to understand your point of view”) were rated on a 5-point scale from 1 (disagree a lot) to 5 (agree a lot). Adequate validity and reliability have been established in prior work (Barnes & Olson, 2003). Children completed a 7-item (α=.77, e.g., “Your parent spent time regularly with you) family routine scale that was adapted from the original Family Routines Inventory (FRI; Jensen et al., 1983), which demonstrated adequate reliability and validity. Parental involvement was assessed using a 9-item scale adapted from Menning (2006) that assessed discrete activities parents can engage in with their children (reliability not calculated because scale measures discrete events; e.g., “During the past month, did you play a sport with your mother/ father?”). Previous work demonstrated predictive validity of the scale for post-divorce father involvement (Menning, 2006). Child monitoring was assessed using a shortened 11-item version of the Child Monitoring Scale (α=.84, e.g., “How much do you know about your child’s use of alcohol?”), which has been found to have good reliability and validity (Hetherington et al., 1992). Finally, closeness was assessed by a single item, “How close do you feel to your mother/father?”, which has been found to predict children’s post-divorce well-being (Stewart, 2003). We conducted a confirmatory factor analysis using a two-factor model separating parental warmth (including involvement, family routines, parent-child communication, closeness, acceptance, rejection) and parental discipline (consistency of discipline and monitoring). The model provided a good fit to the data ($\mathrm{{\rm X}}$²[17] = 52.576, RMSEA = .064, CFI = .976, SRMR = 0.032). We elected to use the parental warmth factor score because there is a strong theoretical rationale to support warmth, or parent-child relationship quality, as the facet of parenting that may drive a protective effect in the relation between exposure to IPC and fear of abandonment (Ainsworth et al., 1978; Hetherington et al., 1992; Wolchik, West, et al., 1993).

Fear of Abandonment (child-report).

At pretest and posttest, children completed 4 true/false items assessing their problematic beliefs about parental separation (e.g., “I worry that I will be left all alone;” “It’s possible my parents will want to live without me”) from the Children’s Attitudes Toward Parental Separation scale (CAPSI; Kurdek & Berg, 1987). Reliability was assessed in a confirmatory factor analysis because the original items had dichotomous response options (Bollen, 1989; Hayduk, 1987). The average reliability (squared correlation of the observed variable and its latent variable) was .75, with factor loadings between .77-.91. Previous work demonstrated predictive validity of the scale for children’s adjustment after divorce (Kurdek & Berg, 1987; Wolchik, Ramirez, et al., 1993; Wolchik et al., 2002).

Child Mental Health Problems (child-, parent-, and teacher-report).

Child internalizing and externalizing problems at the pretest and 10-month follow-up were assessed using the Brief Problem Monitor (BPM; Achenbach et al., 2011), based on child report (19 items; α=.85, α= .88, respectively), parent report (19 items; α=.90, α= .89, respectively), and teacher report (18 items; α=.89, α= .90, respectively). Items are answered on a scale from 0 (not at all true) to 2 (very true). Example items include: “too fearful or anxious” and “disobey my parents/disobedient at home” for internalizing and externalizing problems, respectively. The validity of the BPM is well-established (see Achenbach et al., 2011).

Demographic information.

Parents provided demographic information on themselves and their children (e.g., age, income, time spent with child). Mothers and fathers reported on how many overnights they spent with their child based on the following item: “In the last 30 days, how many nights did your child sleep at your home?” Overnights are frequently used as a measure of parenting time in divorced families (e.g., O’Hara, Sandler, Wolchik, Yun-Tein, et al., 2019; Sandler et al., 2013; Smyth, 2005).

Covariates.

We included child age and child gender as covariates in statistical analyses to account for variance explained by these factors extraneous to our primary research questions. We also controlled for intervention condition because the current study did not focus on intervention effects. In the mediation and moderation models, we included autoregressive controls of dependent variables, including pretest score on fear of abandonment and child internalizing and externalizing problems. In one post-hoc analysis (described below), we included the number of overnights the child spent with the program parent as a covariate.

Data Analysis Approach

Mplus Version 8.3 (Muthén & Muthén, 1998-2017) was used for all analyses. Because the fear of abandonment variable had a skewed distribution with discrete bins, we treated it as a count dependent variable and used a negative binomial hurdle model to estimate A paths in the mediation models, as shown in Figures 2–3 (Hu, Pavlicova, & Nunes, 2011). Hurdle models, similar to zero-inflated models, return two sets of parameters for the count variable. The zero, or hurdle, portion of the model specifies the process that generates either a zero or positive value; in other words, whether the predicted value of the dependent variable is 0 or 1. The count portion specifies the process that generates the range of positive values after the “hurdle” is crossed, which is governed by a model that is truncated at zero and follows a negative binomial distribution (Mullahy, 1986). Hurdle models are distinguished from zero-inflated Poisson (ZIP) models primarily in the way that they treat zero scores. ZIP models differentiate zero scores into two unobserved latent classes, sampling zeros (i.e., scores that could be zero simply due to sampling variability) and structural zeros (i.e, scores that represent a true absence of the phenomenon being measured), whereas hurdle models treat all zeros the same (Loeys, Moerkerke, De Smet, & Buysse, 2012). Because we did not assume different possible meanings of zero counts, we opted to use a hurdle model over a ZIP model. Because there are no distributional assumptions for an independent variable, the B paths (in Figures 2–3) were specified using the fear of abandonment raw scores as the predictor variable.

Figure 2.

Unstandardized regression weights observed in fitted mediation model for child-report mental health problems; IPC = interparental conflict, FoA = fear of abandonment; BPM = Brief Problem Monitor; INT = internalizing problems; EXT = externalizing problems; CI = confidence interval. Dashed lines indicate non-significant path; solid lines indicate significant path.

Figure 3.

Unstandardized regression weights observed in fitted mediation model for teacher-report mental health problems; IPC = interparental conflict, FoA = fear of abandonment; BPM = Brief Problem Monitor; INT = internalizing problems; EXT = externalizing problems; CI = confidence interval. Dashed lines indicate non-significant path; solid lines indicate significant path.

Models were estimated using full information maximum likelihood (FIML) robust standard error correction estimator for unbiased estimates with missing data (Yuan & Bentler, 2000) and the Monte Carlo numerical integration algorithm with 500 integration points. Monte Carlo integration is required for negative binomial hurdle models. According to the Mplus user’s guide Version 8, this is due to use of maximum likelihood estimation and categorical outcomes with a model that has continuous latent variables and missing data. Because children were clustered within families, we used the Mplus sandwich estimator (i.e., command TYPE = COMPLEX) to account for clustering effects in all analyses. In all models, exogenous variables were allowed to covary. Data were missing for child-reported IPC and fear of abandonment at pretest for <1% and fear of abandonment at posttest for 22.9%. At pretest, data were missing for 9–10% of child-, 1% of parent-, and 49% of teacher-reported internalizing and externalizing problems; at the 10-month follow-up, data were missing for 27% of child-, 18% of parent-, and 47% of teacher-reported internalizing and externalizing problems. Overall, of the 559 participants, we had child-, parent-, and teacher-report of child internalizing and externalizing problems for at least one time point (pretest or 10-month follow-up) for 98.4%, 99.6%, and 67.1%, respectively.

We identified two potentially influential cases using regression diagnostics to assess the observed change in the coefficients of theoretical predictors when the cases were excluded (using cutoffs proposed by Cohen, Cohen, West, & Aiken, 2003) and Cook’s distance (using a cutoff of >.20; Bollen & Jackman, 1985). We re-ran analyses without these cases and observed no meaningful change in the results; thus, we reported results with all cases included.

We first tested a longitudinal mediation model in which pretest IPC predicted posttest fear of abandonment (controlling for pretest fear of abandonment), and in turn, posttest fear of abandonment predicted internalizing and externalizing problems at 10-month follow-up (controlling for pretest internalizing and externalizing problems, respectively). We fitted separate models for child-, parent-, and teacher-report of internalizing and externalizing problems. We estimated confidence intervals based on the distribution-of-the-product method (e.g., A₁ x B₁ in Figure 2) to assess the significance of the indirect effects, using the RMediation package (Tofighi & MacKinnon, 2011). We then examined whether parent-child relationship quality moderated the effect of IPC on fear of abandonment by adding the direct effect of relationship quality, as well as the IPC x relationship quality interaction term. We conducted post-hoc probing of significant interactions in several ways, First, we plotted the interaction for illustration and estimated simple slopes of IPC on fear of abandonment at various levels of relationship quality (e.g., at every 10^th percentile and their midpoint when necessary). Then, we conducted a region of significance on X test to identify the values of predictor (i.e., IPC) for which differences in fear of abandonment based on the moderator (i.e., relationship quality) were significant. Finally, we quantified the portion of the sample differentially affected by relationship quality (PA index; Roisman et al., 2012) to better describe the nature of the interaction in terms of dominant models of risk and development, such as diathesis-stress (e.g., risk factors are most likely to impact children who are particularly vulnerable) and differential susceptibility (e.g., vulnerability factors can amplify maladaptation under certain circumstances and promote positive adaptation under others). The PA index is calculated by: (1) identifying the value on the predictor variable at which the simple slope regression lines cross, and (2) calculating the proportion of cases in the sample that have a value higher than the crossover value. This practice is recommended by Roisman et al., 2012 as an adjunct to simple slopes analyses due to their overreliance on null hypothesis significance testing. In cases where the interaction term was nonsignificant, we removed it and tested additive effects. Finally, we used a multiple group procedure to test whether the interaction between IPC and relationship quality on fear of abandonment differed for children whose mothers participated in the program (n = 358) and children whose father participated in the program (n = 201). We used the MODEL CONSTRAINT command in Mplus to test differences in the path coefficients of the effects of IPC, relationship quality, and their interaction.

Results

Preliminary Analyses

Because we were not interested in studying the effects of the two separate intervention conditions, we used a multiple-group procedure with restricted maximum likelihood estimation to test the equivalence of the variance/covariance matrix. We performed a chi square difference test with the Satorra-Bentler scaling correction and did not find significant differences for the child, parent, or teacher models (S-B χ² [45] = 34.935, p = .860; S-B χ² [45] = 56.925, p = .109; S-B χ² [45] = 50.277, p = .272, respectively). We concluded that the parameters (i.e., regression paths, residuals, and variances) were equivalent and thus, combined the two conditions for analyses. Intervention condition was included as a covariate in all models.

Descriptive statistics and zero-order Pearson product-moment correlations are displayed in Table 1. IPC at pretest was positively correlated with posttest fear of abandonment, as well as child- and parent-report of pretest internalizing and externalizing problems. IPC was negatively correlated with parent-child relationship quality at pretest. Posttest fear of abandonment was significantly positively correlated with all reports of mental health problems at all time points, with the exception of pretest teacher-report.

Table 1.

Descriptive Statistics for All Study Variables

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17
1. T1 IPC	-
2. T3 IPC	.49	-
3. T1 RQ	-.32	-.21	-
4. T1 FoA	.25	.17	-.31	-
5. T2 FoA	.23	.17	-.22	.52	-
6. T1 INT child	.34	.28	-.26	.35	.35	-
7. T1 EXT child	.28	.22	-.31	.19	.19	.43	-
8. T1 INT parent	.11	.09	-.01	.14	.21	.33	.08	-
9. T1 EXT parent	.09	.11	-.19	.17	.21	.20	.30	.50	-
10. T1 INT teacher	-.07	-.01	.08	.10	.10	.14	-.02	.29	.16	-
11. T1 EXT teacher	-.11	.06	.03	.01	.03	.02	.08	.03	.31	.26	-
12. T3 INT child	.26	.42	-.23	.29	.32	.56	.25	.20	.11	.11	.05	-
13. T3 EXT child	.06	.23	-.15	.10	.17	.25	.54	.05	.19	.06	.20	.48	-
14. T3 INT parent	.08	.13	-.09	.22	.17	.27	.11	.60	.31	.29	.01	.37	.15	-
15. T3 EXT parent	-.01	.02	-.10	.16	.15	.07	.23	.32	.61	.14	.19	.12	.29	.41	-
16. T3 INT teacher	.04	.01	-.08	.21	.26	.14	.11	.14	.12	.35	.18	.21	.11	.18	.09	-
17. T3 EXT teacher	.00	-.02	.00	.10	.29	.17	.28	.15	.27	.29	.44	.18	.32	.07	.25	.26	-
Mean	0.00	0.01	0.00	0.09	0.10	3.37	3.14	3.31	3.74	1.49	0.87	3.02	3.07	2.49	2.80	1.29	1.08
SD	0.45	0.51	0.76	0.20	0.22	2.65	2.36	2.79	3.07	1.94	1.86	2.93	2.38	2.49	2.69	1.73	1.94

Note. T1 = pretest; T2 = posttest; T3 = 10-month follow-up; IPC = interparental conflict; RQ = parent-child relationship quality; FoA = fear of abandonment; INT = internalizing problems; EXT = externalizing problems; child = child self-report; parent = parent report; teacher = teacher report; SD = standard deviation. Bolded correlation coefficients indicate statistical significance at p < .05.

Does fear of abandonment mediate the relation between IPC and mental health problems?

Child self-report of mental health problems.

Pretest IPC was not significantly related to whether or not children reported any fear of abandonment (i.e., 0 vs. 1) at posttest (p =.158) (See Figure 2). However, pretest IPC was significantly positively related to the severity (i.e., count) of posttest fear of abandonment (b = .659, SE = .285, p = .021, 95% CI = .100, 1.218), after controlling for pretest fear of abandonment and other covariates. In turn, higher fear of abandonment was significantly related to more internalizing problems at 10-month follow-up, after controlling for pretest internalizing problems (b =.569, SE =.204, p = .005, 95% CI =.169, 0.969). The indirect effect of exposure to IPC on internalizing problems through fear of abandonment was significant (a₁b₁ = .372; 95% CI = .033, .858). Posttest fear of abandonment was marginally positively related to externalizing problems at 10-month follow-up (p =.080). Also, there was a direct negative effect from pretest IPC to externalizing problems at 10-month follow-up, after controlling for pretest externalizing problems (b = −.537, SE = .249, p = .031, 95% CI = −1.026, −.048). This effect was in the opposite direction of what was expected; higher exposure to IPC was related to fewer externalizing problems 10 months later.

We conducted post-hoc analyses to understand the unexpected direct effect from pretest IPC to externalizing problems at the 10-month follow-up. First, we examined change in the CPIC, CIM, and BDM scales (which comprise the IPC latent variable) by testing the difference in the means at pretest and the 10-month follow-up; these scores significantly decreased over time (CPIC: pretest M = 1.65, 10-month follow-up M = 1.43, Wald [1] = 72.74, p < .001; CIM: pretest M = 1.92, 10-month follow-up M = 1.83, Wald [1] = 5.51, p = .019; BDM: pretest M = 1.59, 10-month follow-up M = 1.50, Wald [1] = 3.79, p = .052). Second, we examined the unconditional bivariate correlation between pretest IPC and externalizing problems at 10-month follow-up; this relation was small and non-significant (r =.065, p = .217). Third, we added IPC at the 10-month follow-up as an additional predictor in the model to understand the direct longitudinal effect while also accounting for concurrent IPC. In this model, the direct effect from pretest IPC to externalizing problems at the 10-month follow-up remained significant and negative (b = −1.008, SE = .272, p < .001, 95% CI = −1.542, −.474), but there was also a significant effect of concurrent IPC in the expected positive direction (b = .925, SE = .236, p < .001, 95% CI = .463, 1.388), indicating that concurrent level of IPC is an important correlate of externalizing problems, above and beyond IPC assessed 10-months earlier. The path from posttest fear of abandonment to internalizing problems at the 10-month follow-up did not substantively change when concurrent IPC was included.

Parent-report of mental health problems.

The A paths are identical to the child-report model; thus, only B paths are reported here. Posttest fear of abandonment was not significantly related to internalizing (p =.242) or externalizing (p = .348) problems at the 10-month follow-up. Consistent with the child-report model, there was a direct, negative effect from pretest IPC to parent-reported externalizing problems at the 10-month follow-up, after controlling for pretest externalizing problems (b = −.542, SE = .239, p = .023, 95% CI = −1.010, −.074).

As with the child-report model, we conducted post-hoc analyses to understand this unexpected direct effect. We found that the unconditional bivariate correlation between pretest IPC and externalizing problems at the 10-month follow-up was small and non-significant (r = −.007, p = .886). When we added IPC at the 10-month follow-up as an additional predictor, the direct effect from pretest IPC to externalizing problems at the 10-month follow-up was no longer significant (b = −.532, SE = .282, p = .059, 95% CI = −1.086, 0.021). The effect of concurrent IPC on externalizing problems was also not significant (b = .074, SE = .257, p = .773, 95% CI = −.429, .577).

Teacher-report of mental health problems.

The A paths are identical to the child-report model; thus, only B paths are reported here. Posttest fear of abandonment was significantly related to internalizing (b = .514, SE = .181, p = .005, 95% CI = .159, .869) and externalizing (b = .589, SE = .222, p = .008, 95% CI = .155, 1.023) problems at the 10-month follow-up, after controlling for pretest problems (See Figure 3). Both indirect effects of exposure to IPC on internalizing and externalizing problems through fear of abandonment were significant (a₁b₁ =.348; 95% CI = .029, .823; a₁b₂ = .376; 95% CI = .037, .832, respectively). Direct effects of pretest IPC on internalizing and externalizing problems at the 10-month follow-up were not significant.

Does parent-child relationship quality moderate the link between IPC and fear of abandonment?

In the model in which parent-child relationship quality was tested as a moderator of the relation between IPC and fear of abandonment, the interaction term (IPC x relationship quality) was not significant in predicting either the odds of endorsing any fear of abandonment (p = .753) nor the severity of fear of abandonment (p = .726). When we removed the interaction term and tested IPC and parent-child relationship quality as independent predictors, IPC was positively related to the severity of fear of abandonment (b = .652, SE = .286, p = .022, 95% CI = .092, 1.212) but parent-child relationship quality was not (b = −.028, SE = .115, p = .805, 95% CI = −.253, .197). Neither significantly predicted the odds of endorsing any fear of abandonment.

When we used a multiple-group approach to test differential moderation, the interaction term was variant across parent gender (b_diff = 1.790, SE =.783, p = .022, 95% CI =.256, 3.325). The interaction between IPC and relationship quality on fear of abandonment was significant for fathers, but not for mothers. Upon decomposing the interaction, we found that the simple slope of IPC to fear of abandonment severity was positive and significant (b = 1.663, SE = .480, p = .001, 95% CI =.705, 2.621) only for fathers with high (+1 SD) relationship quality. For those with low- and moderate-quality father-child relationships, IPC was not significantly related to fear of abandonment (see Figure 4). Follow-up simple slopes tests revealed that IPC predicted higher fear of abandonment severity for children who reported father-child relationship quality at or above approximately the 60^th percentile in our sample. Conversely, IPC was not related to fear of abandonment for children who reported father-child relationship quality as low as the 5^th percentile in our sample. The PA index showed that 43.3% of cases fell above the point on IPC at which the simple slope regression lines intersect (crossover point = 0.120); representing the proportion that was differentially affected by father-child relationship quality with regard to the relation between IPC and fear of abandonment. Roisman et al, 2012 suggest that PA values above 16% indicate strong evidence for data that are consistent with differential susceptibility theory. The region of significance on X (IPC) test indicated that the differences in fear of abandonment based on father-child relationship quality was significant below IPC = −0.1696 (34.8^th percentile) and above IPC = 0.7906 (94.7^th percentile). We further explored this interaction to test whether relationship quality was operating as a proxy of number of overnights spent with fathers. In the model in which number of overnights was included in place of relationship quality, the interaction term was not significant (p = .355). When we added the number of overnights the child spent with the program parent as a covariate in the original model, the interaction between relationship quality and IPC on the severity of fear of abandonment became marginal (b = 1.244, SE = .687, p = .070, 95% CI = −.103, 2.591); however, the simple slopes were analogous to those in the original model (+1 SD: b = 1.407, SE = .574, p = .014, 95% CI = .281, 2.533), indicating that for children who reported a high quality father-child relationship only, higher IPC predicted fear of abandonment.

Figure 4.

Simple slopes for interaction effect in father subsample. The displayed range of IPC is +/− 2 SD. FOA = fear of abandonment; IPC = interparental conflict; RQ = relationship quality; low father-child RQ = −1 SD; moderate father-child RQ = mean; high father-child RQ = +1 SD. Shaded areas denote regions where the simple slope regression lines are significantly different. Bolded percentages display the proportion of the sample that fell into the regions of significance on IPC. Plotted using the online calculator (http://www.quantpsy.org/interact/mlr2.htm) provided by Preacher et al., 2006.

Discussion

The major contribution of this study is to provide prospective longitudinal support for the theoretical notion that fear of abandonment is a pathway through which exposure to post-divorce IPC leads to children’s mental health problems. Fear of abandonment mediated the effects of IPC on both child- and teacher-reported internalizing problems and teacher-reported externalizing problems. We did not find support for the hypothesis that a high-quality parent-child relationship buffered the adverse effect of IPC on children’s fear of abandonment. However, the link between IPC and fear of abandonment was moderated by father-child but not mother-child relationship quality. We discuss how complex associations among IPC, fear of abandonment, parenting, and child outcomes in divorced families might explain these results.

There are several methodological strengths of the study that bolster confidence in the findings. First, the three-wave longitudinal design provided a robust test of the mediation models. Second, the use of multiple reporters of child mental health problems diminished the likelihood that the results are explained by shared method variance. Third, the sample was relatively large and diverse in terms of ethnicity, child age, and parent gender, which supported the generalizability of the findings.

The finding that fear of abandonment mediated the link between IPC and children’s mental health problems is relevant to two prominent theoretical frameworks about how IPC exerts its negative effects on children’s outcomes. According to the Emotional Security Theory (Cummings & Davies, 2010), children’s response to IPC involves internal representations about how conflict may negatively affect their family’s future (Davies, Forman, Rasi, & Stevens, 2002). Although not explicitly included in the standard measures of emotional security, fear of abandonment has face validity as one type of internal representation that may be linked to IPC. The current findings indicate that for children from divorced families, fear about whether their parents will continue to care for them is a particularly critical internal representation about how IPC will affect them. Future research might examine whether fear of abandonment mediates the effects of IPC on mental health problems in children from married families as well.

The mediational link between IPC and children’s mental health problems through fear of abandonment is also consistent with the Cognitive Contextual Framework (CCF; Grych & Fincham, 1990), which highlights the role of children’s threat appraisals in response to IPC. Further, it aligns with prior literature demonstrating that other types of threat appraisals (e.g., worrying that they will become involved in the conflict, dreading potential harm to themselves or their parents) partially explain the deleterious effect of IPC on mental health problems (Fosco & Feinberg, 2015; Grych, Harold, & Miles, 2003). Although the early conception of the CCF included children’s fears about the stability of their relationship with one or both parents (Grych & Fincham, 1990), this aspect of threat is not explicitly measured in the corresponding instrument (i.e., threat subscale of CPIC; Grych, Seid, & Fincham, 1992). Similar to prior work on emotional security, cognitive appraisal mediational models have been investigated predominantly in children from married or cohabitating families. It is plausible that the content of children’s threat appraisals depends on the context within which IPC occurs. For example, in married families, children may focus on family stability (i.e., whether or not their parents will divorce) or self-blame, whereas when IPC occurs within the context of a divorced family, children may be more likely to focus on threats to the security of their relationship with their parents. Future research focused on understanding whether there are differences in the nature and role of threat appraisals of IPC in married and divorced families would be useful.

The finding that fear of abandonment mediated the association between IPC and both child- and teacher-report of child mental health problems increases confidence that the findings are not due to shared method variance. Teachers were not aware of levels of IPC nor children’s fear of abandonment, so their reports are not confounded by this knowledge. However, the analyses using parent-report of child mental health problems did not support the hypothesized model. It is possible the level of contact with the child or the parent’s divorce-related psychological distress may have affected the parent’s awareness of their child’s mental health problems. Low rates of agreement between parent, teacher and child reports of mental health problems have been consistently documented (Achenbach, McConaughy, & Howell, 1987; De Los Reyes & Kazdin, 2005) and one study showed that parental separation contributed to higher rates of parent-child report discrepancy (Van Roy, Groholt, Heyerdahl, & Clench-Aas, 2010). It is important for future research to explore how best to integrate information from multiple informants when assessing child mental health problems in divorced families.

Contrary to our prediction, parent-child relationship quality was not protective in the link between IPC and fear of abandonment. The finding for mothers is not completely unexpected given the findings of previous studies with divorced samples that were conducted with families with primary maternal custody. Wolchik and colleagues (2002) found only a marginally significant interaction effect where mothers’ quality of parenting mitigated the effect of multiple divorce-related stressors, including IPC, on children’s fear of abandonment. In another sample, Lutzke et al. (1996) found that high-quality mother-child relationships did not protect against mental health problems associated with exposure to IPC. Studies with married families, however, have shown that high-quality parenting mitigates the effects of IPC on mental health problems (El-Sheikh & Elmore–Staton, 2004; Katz & Gottman, 1997; Rutter, 1971). It may be that in the context of divorce, the associations among IPC, parent-child relationships, and children’s outcomes are much more complicated than in married families. Sandler and colleagues (2008) found that for high-IPC divorced families, but not low-IPC divorced families, a high-quality relationship with one parent compensated for the lack of a high-quality relationship with the other parent to protect against internalizing problems. In a different sample, comprised only of divorced families with high levels of IPC, the buffering effect of both mothers’ and fathers’ parenting on children’s mental health problems similarly depended on the quality of the parenting provided by the other parent (i.e., children in high-IPC families benefited from high-quality parenting when parenting by the other parent was low, but when the quality of either parents’ parenting was high, there was no added benefit from high-quality parenting by the other parent; Sandler et al., 2013). Thus, it is possible that a buffering effect would have been found if parent-child relationship quality of both parents were included in the model.

There were two unexpected findings that warrant further discussion. First, in the mediation models, we found a cross-reporter (child- and parent-report) negative direct effect from pretest IPC to externalizing problems at 10-month follow-up. This finding was surprising because decades of literature link IPC to higher externalizing problems (Amato, 2001). There are a couple of plausible explanations. First, this finding may be a statistical artifact. The direct and indirect effects are in opposite directions (i.e., inconsistent mediation) and the lack of significant bivariate correlations between pretest IPC and externalizing problems at 10-month follow-up suggest that fear of abandonment may be operating as a suppressor variable. Because the current study design cannot speak to causal effects, and mediation and suppression models are statistically equivalent, the distinction can only be made on theoretical grounds (MacKinnon, Krull, & Lockwood, 2000). Alternatively, it is possible that this finding should be considered in the context of changes in IPC over time. The prospective paths tested were autoregressive, meaning that the effects represent change in externalizing problems above and beyond the pretest level. Prior research reliably demonstrates that for most families, IPC decreases as a function of time since the divorce (e.g., O’Hara et al., 2019). This pattern was observed in the current study; the mean level of IPC decreased between pretest and the 10-month follow-up. It is possible that children who experienced high initial levels of IPC that decreased over time demonstrated a concomitant spike in externalizing problems initially that lessened over time. Thus, a relative decrease in IPC from pretest to 10-month follow-up might explain why a high pretest level of IPC predicted a decrease in externalizing problems over time. This explanation aligns with the work of many scholars (e.g., Hetherington et al., 1982; Long et al., 1988; Morrison & Coiro, 1999) who have suggested that divorce in high-conflict marriages can be beneficial for children if there is a reduction in exposure to IPC. Further, post hoc analyses demonstrated that concurrent conflict was significantly positively related to child-reported externalizing problems at 10-month follow-up (i.e., the expected direction) and in the parent-report model, the direct effect of pretest IPC on externalizing problems at the 10-month follow-up was no longer significant when concurrent conflict was included, highlighting the importance of considering concurrent IPC.

The other unexpected finding is that the relation between IPC and fear of abandonment was moderated by father-child but not mother-child relationship quality. IPC was positively related to fear of abandonment for all children in the study, except for those who reported low or moderate father-child relationship quality. Children who have particularly close relationships with their fathers have much to lose if those relationships are threatened. Thus, it is not surprising that they report greater abandonment fears under conditions of high IPC. The nature of the interaction is consistent with differential susceptibility theory (Belsky, 1997), indicating that children vary in their susceptibility to post-divorce IPC, both for better and for worse, as a function of the father-child relationship (Roisman et al., 2012). Higher-quality father-child relationships may be beneficial (i.e., relatively lower fear of abandonment if IPC is low) or harmful (i.e., higher fear of abandonment if IPC is high). In this study, a greater proportion of the sample (35%) benefited from a high-quality father-child relationship (in the context of low IPC) than paid a cost for a high-quality father-child relationship in the context of high IPC (5%). We also considered the possibility that this effect may be driven by the degree of contact between the father and child; perhaps children who spent fewer overnights with their father did not fear being abandoned as much as those who had more contact. This hypothesis was based on a prior study which demonstrates the association between positive parenting and child mental health problems is significant only when parents have a minimum number of overnights with their child in cases of high-conflict divorce (Sandler et al., 2013). However, when the number of overnights was included as a covariate in the model testing the interaction between IPC and relationship quality on fear of abandonment, the results did not substantively change. Further, when we replaced parent-child relationship quality with number of overnights in the moderation analysis, number of overnights did not significantly moderate the association between IPC and fear of abandonment.

Implications for Intervention Development

The results of this study suggest two directions for improving prevention programs for divorced families with high IPC. First, the findings highlight the need to optimize coping programs for children to reduce their fear of abandonment. One approach would be to increase children’s capacity to reappraise fear of abandonment in response to IPC. For example, programs could help children view the conflict as less threatening (e.g., “They are angry at each other, not me.” or “They really dislike each other but they still love me.”) It is important to note that many children do in fact spend less time with one or both parents after divorce, which may increase their concerns about their needs not being met. Thus, the goal of interventions is to help a child differentiate realistic and unrealistic fears and tailor cognitive reappraisal strategies to the child’s situation. Another approach is to educate parents about common fears children have, including fear of abandonment, and provide tools to help them remain sensitive to these fears and responsive to their child’s needs, including reassuring the child that they will be there for them. Second, although rigorous evaluations have shown that prevention programs reduce risk for children by improving positive parenting after divorce (Forgatch & Degarmo, 1999; Tein, Sandler, MacKinnon, & Wolchik, 2004; Zhou, Sandler, Millsap, Wolchik, & Dawson-McClure, 2008), the current results indicate that high-quality parenting does not eradicate the deleterious effects of IPC, at least not through its effects on children’s fear of abandonment. Randomized trials of two parenting programs have found encouraging effects to reduce IPC (Cookston, Braver, Griffin, De Luse, & Miles, 2007; Tein et al., 2018). However, these programs are both costly and lengthy, which reduces the likelihood of widespread adoption and implementation. Given that exposure to IPC and high-quality parenting are two of the most robust predictors of children’s mental health problems following divorce, developing brief, scalable programs that both improve parenting and reduce children’s exposure to IPC is needed to reduce the public health burden of parental divorce, particularly for high-IPC families. It will be important for parenting interventions to focus on sensitizing parents to the impact of IPC behaviors, such as frequent and intense arguing, and teach skills to minimize these behaviors. Attention to reducing behaviors that triangulate children to create alliances or undermine the other parent is also important given that these behaviors likely foster fear of abandonment. In addition, this work underscores the importance of legal approaches to dispute resolution for separated/divorcing parents (e.g., mediation [Emery et al., 2005], collaborative divorce [Pruett et al., 2005]) that have the potential to reduce interparental conflict.

Limitations and Future Directions for Research

There are limitations of the current study that should be addressed in future research. First, child report of parent-child relationship quality was only assessed for the parent who participated in the program. Given prior research demonstrating that there may be higher order interactions between mother’s and father’s parenting after divorce (Sandler et al., 2008, 2013; Sobolewski & Amato, 2007), studies that include information about the child’s relationship with both parents would provide a more contextualized picture of the associations among IPC, parent-child relationship quality, fear of abandonment, and mental health problems. Second, there may be important mediators missing from the models. For example, two studies (Fosco & Bray, 2016; Grych et al., 2003) have found that threat appraisals (e.g., concern that a parent will be harmed; ambiguous worry about negative implications of the conflict, perceived inability to cope) predict internalizing problems and self-blame predicts externalizing problems. Because the current study focused on only one type of threat appraisal (i.e., fear of abandonment) it will be important for future research to also assess other types of common maladaptive appraisals, such as self-blame and a sense of responsibility to stop the conflict. Such research is particularly germane to developing tailored interventions that are appropriate for different children at risk for mental health problems (i.e., personalized medicine; Bickman et al., 2016; Shoham & Insel, 2011). Third, data were drawn from a larger study of a parenting-after-divorce program and those who were either receiving mental health treatment or reported highly elevated mental health problems, including suicidal ideation, were excluded. Thus, the current findings may not generalize to the broader population of divorcing and separating families, especially those with severe mental health problems. Fourth, the path analysis model fitted in this study focused solely on between-person differences in levels between waves specified in the model, which precludes interpretations about within-person change in model variables across time. An important future direction is to examine the extent to which these key constructs change over time within individuals as well as how much such change varies from person-to-person. Finally, data on time since physical separation were not collected. Because IPC and fear of abandonment may be higher and parent-child relationship quality lower immediately after the physical separation, it will be important to assess whether the associations between IPC, fear of abandonment, parent-child relationship quality, and mental health problems change across time.