Heterogeneity in the dynamic arousal and modulation of fear in young foster children

Abstract

Guided by emotional security theory, we explored how child and context-related factors were associated with heterogeneity in young foster children’s organized patterns of fear response to distress. Results from group-based trajectory modeling used to analyze observational data from a fear-eliciting task showed that children from our sample (mean age = 62 months, SD = 9) were classified into 3 specific fear regulation patterns differentiated by the emotional response parameters of onset intensity, peak intensity, and rise time. A descriptive examination of child’s emotion knowledge, aggressive behaviors, and attention problems, as well as length of time in current foster home, placement transitions, and caregiver responsiveness and modeling showed class-specific differences in means. Moreover, the likelihood of class membership was significantly predicted by children’s emotion knowledge, aggressive behaviors, and foster mothers’ responsiveness and modeling of appropriate boundaries. Results show promising support for the implementation of individualized, child-directed interventions targeting specific patterns of response parameters of emotion regulation for young foster children. Further, parenting intervention services need to promote the emotion socialization skills of foster parents that are tailored toward each specific trajectory pattern of emotion arousal and modulation.

1. Introduction

In 2017, over 4 million children were referred to Child Protective Services (CPS) agencies across the United States due to allegations of maltreatment, with rates of referrals being disproportionately higher for young children aged 0 through 5 (U.S. Department of Health & Human Services, 2019). Within the same year, more than 127,000 young children entered foster care, with about 35% having experienced placement instability (Administration for Children & Families, 2019). Instability is defined as 3 or more placements (Webster, Barth, & Needell, 2000), with disruptions in care that typically occur within the first 24 months of placement (AUTHOR 3A). Unfortunately, these early experiences of maltreatment and placement instability are associated with negative effects on children’s socioemotional development, particularly within the domain of emotion regulation (Maughan & Cicchetti, 2002; Pollak & Kistler, 2002), with long-term effects on school readiness and later academic achievement (AUTHOR 1).

Emotion regulation is a system of cognitive, behavioral, and emotional processes necessary for monitoring, evaluating, and modifying the intensive and temporal features of emotional reactions. These emotional reactions, in turn, are important for facilitating the transactions between an individual and his or her context to achieve a goal (Cole, Martin, & Dennis, 2004; Thompson, 1994). Children’s capacity to regulate emotions develop from interactions between a parent and child as regulatory skills move from co-regulation to self-regulation (Lunkenheimer, Kemp, & Albrecht, 2013). The ability to regulate fear is particularly important because it has the potential to mitigate the negative effects of maltreatment on future development (Denham, Bassett, & Wyatt, 2007; Hastings & De, 2008). Given the high incidence of maltreatment among young children, as well as the importance of regulating emotions, it is critical to further understand individual differences in patterns of fear regulation. For example, capturing heterogeneity in the dynamic temporal features of young maltreated children’s emotion regulation in general, and fear regulation specifically, may inform our understanding of emotion regulation in this vulnerable population of children. Further, given the importance of caregivers in the development of emotion regulation, it is important to understand how parenting in the context of foster care affects heterogeneity in these regulatory patterns in order to elucidate potential risk and protective factors in the developmental course of children’s regulatory problems.

To address these gaps, we employed a finite mixture modeling framework (Masyn, 2013; Nagin & Odgers, 2012) to explore heterogeneity in the patterns of young foster children’s arousal and modulation of an activated emotion during a fear assessment task. We also examined how child factors (i.e., emotion knowledge and behavior problems that include internalizing and externalizing behaviors) and context factors (i.e., length of time in placement, placement instability, and caregiving quality) affected the likelihood of belonging in specific fear trajectory patterns.

1.1. Emotion regulation and maltreatment

The development of emotion regulation abilities in children are affected by the family context through children’s observation of how emotions are managed, parenting practices related to the socialization of emotions, and the emotional climate of the family that includes attachment and parenting styles (Eisenberg, Cumberland, & Spinrad, 1998; Morris, Silk, Steinberg, Myers, & Robinson, 2007). Within the context of maltreatment, these social relationships may be compromised, which can adversely impact the development of children’s regulatory abilities, particularly when mediated by specific child- and context-related factors such as attentional bias toward negative-valenced emotions (Pollak, Cicchetti, Hornung, & Reed, 2000; Shackman, Shackman, & Pollak, 2007), problems with understanding emotions (Cicchetti & Valentino, 2006; AUTHOR 3B), internalizing and externalizing behavior problems (Kaplow & Widom, 2007; Rogosch, Oshri, & Cicchetti, 2010; Shields & Cicchetti, 2001), problems with parents’ capacity to express positive and negative-valenced emotions (Milojevich & Haskett, 2018), harsh parenting (Chang, Schwartz, Dodge, & McBride-Chang, 2003; Lunkenheimer, Ram, Skowron, & Yin, 2017), and interadult violence (Maughan & Cicchetti, 2002).

For children in foster care, the high incidence of placement instability threatens a young child’s ability to develop a stable and responsive parent-child relationship that is important for mitigating the effects of early adversity, as well as supporting the ongoing development of regulatory abilities (Casanueva et al., 2014). The absence of a stable, positive caregiver has been shown to increase behavior problems in children (Rubin, O’Reilly, Luan, & Localio, 2007), which in turn increases the likelihood of further disrupting placement (Newton, Litrownik, & Landsverk, 2000) that further compromises children’s emotion regulation development. Taken together, child and context factors related to early experiences of maltreatment and placement instability can lead to variability in the development of emotion regulation abilities.

According to Thompson (1990), variability in emotional reactions may index heterogeneity or individual differences in regulatory processes, signaling the need for different ways to study emotional behaviors. For example, a study by Maughan and Cicchetti (2002) found heterogeneity in children’s emotion regulation patterns wherein a majority of maltreated preschool-aged children, compared with less than half of non-maltreated children, exhibited dysregulated patterns of anger that included undercontrolled/ambivalent (U/A) and overcontrolled/unresponsive (O/U) types. These findings are important in providing evidence for variability in overall regulation patterns as a result of early adversity. However, it is less clear if similar patterns of heterogeneity exist in the systematic changes (i.e., temporal features) of an activated emotion, which is key to defining emotion regulation (Cole et al., 2004; Thompson, 1994).

Two important characteristics to consider when studying emotional behaviors include emotional tone and emotional dynamics (Thompson, 1990). The first characteristic, emotional tone, includes specific emotions that characterize positive- or negative-valenced emotional responses. Examples of such discrete emotions include joy, interest, anger, sadness, disgust, and fear. The second characteristic of emotional behaviors, emotional dynamics, consists of “response parameters that define the quality of emotional behavior regardless of its tone, and that often reflects the influence of diverse emotional regulatory processes” (Thompson, 1990, p. 372). These response parameters include onset intensity (intensity of an elicited emotion at the introduction of a stimulus), peak intensity (intensity of an elicited emotion at its peak), and rise time (amount of elapsed time between the initial emotional response and peak intensity). Taken together, these response parameters are indicative of a hypothesized quadratic arc in the arousal and modulation of an elicited emotional response. Evidence of heterogeneity in these intensive and temporal features may provide further understanding of how patterns of emotion regulation foster or undermine effective behavioral organization towards goal attainment.

1.2. Fear response

Fear is an important emotional tone to consider in young children’s emotional behavior, particularly within the context of early adversity. From a functional perspective, fear is a necessary emotional response that is activated in the presence of threat (Rothbart & Bates, 2006). As a broad construct, fear includes several indicators such as motivational aspects of attention toward or withdrawal away from the fear-inducing stimulus, behavioral reactions that include fight, flight, or freeze responses, distress behaviors such as crying or fear-related facial expressions, and social reactions such as social reticence (Buss, 2011; Kagan, 1994).

Preschoolers’ dysregulated fear response has been associated with increased risk for socially anxious behaviors by kindergarten (Brooker, Kiel, & Buss, 2016; Buss et al., 2013) as well as later development of anxiety disorders (Clauss & Blackford, 2012). For children with a history of maltreatment, evidence of hyperactivation in the fear-processing regions across the ventromedial prefrontal cortex and anterior cingulate cortex has resulted in faster responding and bias to fear-related expressions (Hart et al., 2018; Keding & Herringa, 2015; Suzuki, Poon, Kumari, & Cleare, 2015). On the other hand, patterns of hypoactivation have also been seen in youth who experienced early adversity (Gotisha et al., 2014). These contradictory findings, along with the complexity of fear, give rise to the notion that children’s pattern of fear arousal and modulation is not homogenous. Indeed, evidence for heterogeneity in fear regulation was found by Buss (2011) where the transaction between profiles of fear regulation and children’s context at 24 months predicted anxious behaviors by kindergarten. Given that fear organizes children’s regulatory behaviors toward a goal of safety, it is important to understand how these goals develop within the context of foster families and early adversity.

1.3. Emotional security as a goal

One way to understand how dysregulated fear responses emerge, particularly for young children with early maltreatment history, is through the lens of Emotional Security Theory (EST-R; Davies & Martin, 2013). According to EST-R, early home experiences are influential in organizing children’s behavioral responses to emotion-eliciting stimuli (i.e., functional nature of emotions). As a primary human goal, children’s emotional security regulates, and is regulated by, actions and reactions across multiple caregiving relationships (Davies & Martin, 2013; Davies, Winter, & Cicchetti, 2006). The notion of preserving emotional security stemmed from early ethological descriptions of the fear system and is rooted in attachment theory (Davies & Sturge-Apple, 2007). EST-R posits that children’s goal of maintaining emotional security is achieved through their social defense system (SDS). Interpersonal discord, rejection, or hostility from social relationships (e.g., with foster parents) can undermine safety and predictable access to resources, which shapes children’s profiles of security. These early social relationships organize children’s SDS into 4 specific profiles that include secure, mobilizing, dominant, and demobilizing (Davies & Martin, 2013; Davies & Sturge-Apple, 2007).

Children classified as secure exhibit mild to moderate distress commensurate with the intensity of the threat. These children are prone to low distress, are easily soothed, and exhibit high effortful control. Within the family context, parents exhibit high levels of responsiveness, and can model and coach appropriate management of emotions. Next, children classified as mobilizing show blatant displays of vigilance and distress in response to threat. Children in this classification are prone to high distress, difficult to soothe, and exhibit high impulsivity and approach. Parents show moderate levels of responsiveness and psychological control. Third, children classified as dominant exhibit hypervigilance in response to threat. They suppress vulnerable emotions, show more anger and hostility, and respond with reactive forms of aggression such as yelling or hitting. These children are prone to low or moderate levels of distress, exhibit low effortful control, and are highly impulsive. The family environment is typically chaotic with high levels of instability, and parents are typically unresponsive, controlling, and show a high degree of apathy towards children’s emotions. Finally, children classified as demobilizing show freezing, lethargic, or cutoff behaviors (e.g., covering eyes) in response to threat. Children in this classification exhibit low sensitivity to reward, display moderate effortful control, show low levels of impulsivity and high behavioral inhibition, and often exhibit disorganized attachment patterns. Parents are often unresponsive to, and intolerant of, children’s emotions and emotional expressions, with a high potential for parental abuse in these family contexts.

1.4. The current study

Taken together, prior research has shown that early experiences of adversity negatively affect young children’s regulation of fear. However, there is a paucity of research examining fear regulation in young children exposed to maltreatment and foster care placement. Additionally, there is evidence that child and context-related factors are associated with children’s organized patterns of response to distress or threat in order to maintain emotional security, indicating heterogeneity in regulating emotions. Less clear, however, is an understanding of how similar patterns of heterogeneity can be seen in more dynamic models of fear arousal and regulation, and how membership in a specific trajectory class is associated with child or foster care placement factors. Using a finite mixture modeling approach and an observational task to assess fear, we tested the hypothesis that young foster children with a history of maltreatment will exhibit different patterns in the arousal and modulation of fear. More specifically, we explored how heterogeneity in fear regulation, indexed by children’s emotional response parameters, were related to SDS profiles of security. Inclusion of child-level (i.e., emotion knowledge, internalizing behaviors, and externalizing behaviors) and context-level (i.e., length of time in current foster placement, placement transitions, and caregiver responsiveness and acceptance) factors adds to the test of this hypothesis by understanding how these covariates predict the likelihood of belonging in each profile of security.

2. Methods

2.1. Participants

Participants in the study were 45 young children (53% female; M = 62 months, SD = 9 months) placed in foster care (65% placed in non-kinship care) and their foster mothers (M = 47 years, SD = 11.6 years). The children and their foster mothers were recruited from two child welfare agencies in a suburban area within the mid-Atlantic region. All children had a substantiated case of child maltreatment in which 75% experienced neglect, 3% experienced physical abuse, 3% experienced sexual abuse, and 19% experienced multiple types of maltreatment. Most of the children (75%) were 24 months or older at time of first out-of-home placement. The mean length of time in children’s foster placement at the time of the study was 21.21 months (SD = 17.78 months), with a high rate of placement instability (M = 2.93, SD = 1.77). The children were predominantly African-American (75% African-American; 7% White; 2% Hispanic; 11% mixed parentage; and 5% other). Foster mothers were predominantly African- American (64% African-American; 12% White; 2% Hispanic; 2% mixed parentage; and 18% other). Most foster mothers had some college (40% some college or associate degree; 10% college degree; 10% advanced degree) and were employed outside the home (71% working). Median annual income for foster families was $50,000 - $59,000. Over one third of foster mothers were married (38%), whereas more than a quarter were separated or divorced, and 16% reported never being married.

2.2. Procedures

Procedures for the current study were approved by the University of Maryland Institutional Review Board and the Department of Children’s Services responsible for the care of the children provided initial consent for study participation. Informed consent was also obtained from each foster parent participant. Foster mothers were given incentives for their participation and children were given developmentally appropriate gifts.

Data were collected over the course of two sessions, the first in foster families' homes where direct assessments were conducted, and the second in a research laboratory where the structured observational task for the present study was conducted. The fear episode was drawn from the preschool version of the Laboratory Temperament Assessment Battery (Lab-TAB; Goldsmith, Reilly, Lemery, Longley, & Prescott, 1993) using the Scary Mask structured task designed to examine fear responses in preschool-aged children. The task was carried out in a large room with a two-way mirror where children’s responses were video recorded and later coded. Each task session began with the child sitting at a child-sized table. Once the child was situated, the foster mother and a familiar research assistant left the room and waited outside the door. Immediately after, another research assistant (i.e., “a friendly stranger”) entered the room, introduced herself, and attempted to engage the child in a positive conversation for 30 seconds (e.g., “What’s your name? You’re doing really well today!”). The conversation concluded with the friendly stranger saying in a slow and “ominous” tone, “hold on, I have one more thing I’d like to show you.” At this point, the friendly stranger left the room briefly to put on a gorilla mask. She walked back into the room and sat silently to 30 seconds. If a child exhibited severe distress, the task was terminated early and the foster mother and the familiar research assistant were called into the room to help soothe the child. At the conclusion of the task, the friendly stranger removed the mask, invited the child to examine the rubber mask, and asked the foster mother and familiar research assistant to return.

2.3. Measures

2.3.1. Fear arousal and modulation

Children’s elicited fear expressions were scored during the Scary Mask structured task following guidelines from the Lab-TAB manual (Goldsmith et al., 1993). Indicators of children’s fear included an item for facial expressions, an item for distress vocalization, and 4 items for bodily expressions. The observations were scored, in 5-second intervals beginning with the introduction of the scary mask (i.e., baseline) for a total of 25 seconds (i.e., 6 epochs), by two coders trained to reliability (86% agreement on 40% of the sample, κ = 0.79). Facial expressions of fear were coded using the AFFEX system (Izard, Dougherty, & Hembree, 1983) to indicate the intensity of fear expression across three facial regions (see Goldsmith et al., 1993, Appendix A for a definition of AFFEX facial expressions). Distress vocalizations were coded based on intensity of fear such as mild vocalizations, vocalization that indicate some fear (e.g., “who are you?”), or vocalizations that indicate definite fear (e.g., “stay away”). Bodily expressions were coded based on the level of decreased physical activity (e.g., visible tensing of the muscles, freezing with very little motion), avoidance (e.g., turns away, runs to a corner or outside the room), fidget (e.g., movement without a purpose), and startle during the first epoch. For preliminary analyses, the 6 indicators of fear were averaged across all 6 epochs to derive a composite score (Cronbach’s α = 0.71). In order to model temporal changes in fear arousal and regulation, a mean score for the indicators of fear was created for each epoch and employed as a repeated measure for the main analyses.

2.3.2. Emotion knowledge

The ability to understand emotions was measured using the Affect Knowledge Task (Denham, 1986) where children examined pictures that depicted facial expressions of emotions (i.e., happiness, sadness, anger, and fear). During the laboratory visit, the children were presented with the emotional stimuli and asked to explicitly label an emotion (e.g., “how does this person feel?”) or identify the correct emotion depicted by a picture (e.g., “show me the happy face”). Correct responses were summed to create a total score that represented children’s knowledge of emotions, with a possible range of 0 to 64 (Cronbach’s α = 0.89).

2.3.3. Quality of the home environment

The Early Childhood version of the Home Observation for Measurement of the Environment Inventory (EC–HOME; Caldwell & Bradley, 1984) was used to assess the home environment. The EC–Home had 55 items divided across 8 subscales (Cronbach’s α = 0.69) and scored dichotomously as yes or no. The EC–Home was administered via semistructured interviews during the home visit portion of the study. Given the important role that family relationships play in the subjective experiences of security and threat in children’s development of SDS profiles, 2 specific subscales (i.e., responsiveness and modeling) were selected for the present study. The responsiveness subscale included 7 items that examined caregivers’ capacity to be emotionally responsive to children’s concrete, affective, and developmental needs (e.g., “parent holds child close 10-15 min per day,” “parent converses with child at least twice during visit”). The modeling subscale included 5 items that assessed caregivers’ use of appropriate boundaries in the relationship (e.g., “child can hit parent without harsh reprisal”,” “child can express negative feelings without reprisal”).

2.3.4. Behavior problems

Children’s behavior problems were assessed using the Child Behavior Checklist (CBCL/4-18; Achenbach, 1991), which consisted of 113 items (Cronbach’s α = 0.80) rated on a 3-point scale (i.e., very or often true, sometimes or somewhat true, not true) divided across 8 syndrome subscales (i.e., aggressive behavior, anxious/depressed, attention problems, rulebreaking behavior, somatic complaints, social problems, thought problems, and withdrawn). The current study employed the aggressive behavior and attention problems subscales. Completion of the CBCL occurred during the home visit portion of the study.

2.3.5. Child welfare-related information

Child and family background information, including child welfare-related information (i.e., placement transitions and length of time in current placement), were gathered from project-developed questionnaires administered to foster mothers during the home visit portion of the study.

2.4. Analytic Plan

First, in order to confirm that the arousal and modulation of fear within an episode exhibited a quadratic function, unconditional latent growth curve (LGC) models were initially used to rule out other forms such as linear or cubic functions. The usefulness of specifying a quadratic function is often due to its effectiveness in summarizing empirical processes within applied developmental research (Cudeck & du Toit, 2002), particularly given that response variables in developmental processes (e.g., fear expression) are inherently nonlinear (AUTHOR 2). This is especially important when these processes occur within a restricted portion of development (AUTHOR 2), which in this case was the assessment of fear expression within a single episode.

Second, unconditional latent class growth models or group-based trajectory (GBT, Nagin, 1999; 2005) models were employed in order to estimate the number of latent classes based on the dynamic patterns of fear arousal and modulation. GBT models are part of a finite mixture modeling framework where the overall distribution of one or more variables is assumed to be a “mixture of or composite of a finite number of component distributions, usually simpler and more tractable…than the overall distribution” (Masyn, 2013, p. 551). Unconditional GBT models group individuals into their most likely classes based on the posterior probabilities of estimated fear trajectory patterns defined by the quadratic parameters. Unfortunately, these parameters are often difficult to interpret and does not lend itself to interesting comparisons between groups (Cudeck & du Toit, 2002; AUTHOR 2).

To address this limitation, the third step re-specified an alternative form of the quadratic model. According to Cudeck and du Toit (2002), re-expressing the parameters of a quadratic model is possible, particularly when the domain of the independent variable includes a zero (i.e., baseline) and a global maximizer (i.e., rate of change or rise time). The resulting re-parameterized form of the quadratic model is

$$\begin{gathered} g(x_i,\theta )=h_1(x_i,\alpha ) \\ ={\alpha }_y-({\alpha }_y-{\alpha }_0){\left(\frac{x_i}{{\alpha }_x}-1\right)}^2 \end{gathered}$$ (1)

where α₀ is the initial performance at x = 0, α_x is the maximizer of h₁(x_i, α), and α_y is the maximum value of h₁(x_i, α). Applied to a single episode of fear arousal and modulation, these parameters represent baseline (α₀), peak intensity (α_y), and time to reach peak intensity (α_x). See Cudeck and du Toit (2002) for a detailed explanation of the derivation process resulting in these new parameters. These new coefficients now lend themselves to more meaningful interpretation, particularly as they relate directly to the response parameters of emotional dynamics (Thompson, 1990). Figure 1 illustrates how these parameters fit within the hypothetical functional form of fear expression.

Figure 1.

Hypothesized functional form for the arousal and modulation of fear with corresponding response parameters (Cudeck & du Toit, 2002) guided by emotional dynamics (Thompson, 1990).

Finally, conditional GBT models using the manual 3-step process were employed in order to examine the impact of covariates on class membership (Asparouhov & Muthén, 2014; Nagin & Odgers, 2012). Covariates were examined in terms of the likelihood of membership into specific latent classes compared to a reference class by specifying the probability of classification based on a multinomial logit model (Masyn, 2013; Nagin & Odgers, 2012).

2.4.1. Fit and adequacy of the models

Unconditional LGC models were assessed using the joint criteria from Hu and Bentler (1999) including the Standardized Root Mean Squared Residual (SRMR) and Root Mean Square Error of Approximation (RMSEA). This two-index presentation strategy was recommended by Hu and Bentler (1998), who further recommended the use of SRMR and at least one supplemental index, in this case RMSEA (< 0.06), to interpret the fit of models with maximum likelihood estimation methods. Based on their sensitivity analyses, SRMR was the most sensitive to model misspecification but robust to issues of smaller sample sizes (less than 250 participants). The SRMR is a measure of absolute model fit and captures the difference between the observed and model-predicted correlation, without penalizing more complex models with additional paths. Simulation research suggests that an SRMR of less than .08 indicates good model fit (Hu & Bentler, 1999).

Unconditional GBT models were assessed using the sample size adjusted Bayesian Information Criteria (S-BIC; smaller value indicates better relative fit), the Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMR-LRT; low p-value indicates rejecting the k – 1 class model for the k class model), and entropy (> 0.80 indicates high confidence in classification and adequate class separation) (Nylund, Asparouhov, & Muthén, 2007; Ram & Grimm, 2009). Additional assessment followed guidelines set forth by Nagin and Odgers (2012). These included the average posterior probability (AvePP; > 0.70), the Odds of Correct Classification (OCC; > 5.0), and the difference (in absolute terms) between the estimated probability of class membership and the proportion of individuals assigned to that class based on the posterior probability (∣% Diff∣; < 50%).

2.4.2. Sample size and power

Growth models for the current study were used in an exploratory manner to extract meaningful and interpretable parameters to understand developmental processes from the rich repeated measures data. The uniqueness of the foster care sample and application of novel methods in this manner can help provide a descriptive understanding of regulatory patterns in the dynamic arousal and modulation of fear. However, we do not claim that subsequent results are evidence for statistical inference and that such exploratory results require further confirmatory analyses.

A study by MacNeill, Ram, Bell, Fox, and Perez-Edgar (2018) employed both linear and non-linear growth models using EEG data modeled across 7 repeated observations for 28 infants and showing model convergence despite the small sample. Within a mixture modeling framework, Masyn (2013) indicated that sample size may not necessarily have concrete guidelines since adequate fit depends on model complexity, properties of the latent class indicators, and the number, nature and separation of the true classes. This means that satisfaction of the model fit and adequacy assessments discussed above should indicate adequacy of the sample size. Additionally, Ram and Grimm (2009) provided an illustrative example using a more complex 2-class growth mixture model that yielded adequate fit with as little as 34 participants. Taken together, these studies provide evidence that the current study’s sample size should be sufficient for model convergence across the analytic steps outlined earlier. Analyses for the present study were run in Mplus 8.3 using an MLR estimator.

3. Results

3.1. Preliminary analyses

Prior to addressing the study’s main objectives, preliminary analyses were conducted to examine the distribution of fear expression and its correlation with study variables. Average fear expression was represented as a univariate outcome in this preliminary stage by calculating the mean score of fear across all epochs. There were no significant differences in average fear expression by gender, t(43) = −1.52, p = 0.14 and placement type, t(41) = −1.53, p = 0.14 and were therefore excluded in subsequent conditional models. Descriptive statistics and correlations across study-specific variables are presented in Table 1. Child emotion knowledge and caregiver responsivity were the only covariates that significantly correlated with average fear expression (r = −0.32, p < 0.05 and r = −0.31, p < 0.05 respectively).

Table 1.

Correlations and Descriptive Data across the Overall Sample (n = 45)

	1	2	3	4	5	6	7	M	SD
1. Child fear expression	1							0.27	0.17
2. Child emotion knowledge	−0.33*	1						50.29	11.12
3. Child aggressive behaviors	0.07	0.28	1					10.71	7.62
4. Child attention problems	0.05	0.08	0.58***	1				3.67	3.55
5. Length of time in current placement (in months)	−0.10	0.15	−0.09	−0.10	1			21.21	17.78
6. Placement transitions	0.13	0.02	0.16	−0.01	−0.45**	1		2.93	1.77
7. Caregiver responsiveness	−0.31*	0.03	−0.01	−0.17	−0.03	0.08	1	6.00	0.99
8. Caregiver modeling	0.01	0.23	−0.14	−0.18	0.34	0.06	0.6	3.98	0.80

^*p < .05

^**p < .01

^***p < .001

Further examination of the distribution of average fear scores showed a pattern of nonnormality (skewness = 1.34, kurtosis = 2.04), which may be attenuating these bivariate relationships (Bishara & Hittner, 2015). This pattern of univariate non-normality, however, may represent a mixture of two or more normally distributed meaningful subpopulations (McLachlan & Peel, 2000) that represent heterogeneity in the developmental process of interest. According to Masyn (2013), variability in measured outcomes is particularly important given that direct applications of finite mixture models require the a priori assumption that there is heterogeneity in the overall population comprised of a finite number of meaningful subpopulations. The next step, therefore, proceeded with the a priori assumption that heterogeneity in the dynamic pattern of fear regulation existed by first modeling the functional forms of fear arousal and modulation, then estimating GBT models that represented this heterogeneity.

3.2. Modeling the functional form of fear arousal and modulation

Functional forms of fear regulation, represented as linear, quadratic, and cubic growth models, were fit one at a time to the fear expression scores across the 6 epochs. The linear growth model did not yield an acceptable fit (SRMR = 0.36; RMSEA = 0.40) whereas the quadratic model yielded adequate model fit (SRMR = 0.09; RMSEA = 0.06). The cubic growth model did not converge and was therefore excluded from further consideration. Based on the model fit criteria and hypothesized dynamic within-task trajectory of fear arousal and modulation (see Fig. 1), the quadratic model was retained for subsequent analyses. Parameter estimates for the retained model indicated that the quadratic curvature was in the hypothesized direction (intercept = 0.09, p < 0.001; slope = 0.41, p < 0.001; quadratic = −0.08, p < 0.001). Next, in order to test the hypothesis that dynamic patterns of fear expression can be grouped into heterogenous SDS profiles of emotional security, unconditional GBT models were estimated.

3.3. Estimating classes based on dynamic patterns of fear arousal and modulation

Results indicated that the 2- and 3-class models fit the data well, while the 4-class solution did not yield an acceptable model fit (see Table 2). Given that all the children in the sample had a substantiated case of maltreatment, the hypothesized secure SDS profile may not be present in the sample. Therefore, the 3-class model was retained to correspond with the 3 remaining SDS profiles. Further assessment of the adequacy of the 3-class model yielded good fit (see Table 3), indicating the appropriateness of this model. Within this 3-class model (see Fig. 2), the largest group of children (69%; n = 31) were classified as exhibiting a pattern of low intensity of fear expression (intercept = 0.05, slope = 0.26, quadratic = −0.04), 18% (n = 8) were classified as exhibiting a pattern of medium intensity (intercept = 0.19, slope = 0.56, quadratic = −0.11), and 13% (n = 6) were classified as exhibiting a pattern of high intensity (intercept = 0.12, slope = 0.69, quadratic = −0.10).

Table 2.

Unconditional Model Comparison for Number of Latent Trajectory Classes for Fear Arousal and Modulation

Trajectory Class	VLMR-LRT	Entropy
1	--	--
2	96.51**	0.98
3	48.85**	0.97
4	0.42	0.90

Note: VLMR-LRT = Vuong-Lo-Mendell-Rubin Likelihood Ratio Test.

^*p < .05

^**p < .01

^***p < .001

Table 3.

Additional Assessment of Model Adequacy for the 3-Class Model

Trajectory Class	AvePP	OCC	∣% Diff.∣
1 High fear intensity	0.99	2138.97	0.86
2 Medium fear intensity	0.99	344.77	1.51
3 Low fear intensity	0.99	65.22	0.56

Note: Fit criteria: AvePP: >0.7, OCC: > 5, ∣% Diff∣: < 50%

Figure 2.

Estimated sample means for the 3-class model with patterns of low (n = 31), medium (n = 8), and high (n = 6) intensity of fear expresion.

3.4. Re-parameterization of the model: Response parameters of emotional dynamics

Table 4 includes a summary of the class-specific response parameters that include onset intensity, rise time, and peak intensity. Results indicate that children classified as exhibiting a pattern of medium fear intensity had the highest baseline or onset intensity of fear at the time the mask was introduced and was the fastest to reach peak intensity compared with the other 2 groups. Children classified as exhibiting a pattern of high fear intensity displayed the highest peak intensity but took the longest to achieve this level of intensity. Finally, children classified as displaying a pattern of low fear intensity showed the lowest intensity of fear at the onset and peak. To further understand how these fear responses emerged, the next step examined child and contextual factors as covariates in the conditional model.

Table 4.

Class-Specific Response Parameters for the Dynamic Arousal and Modulation of Fear Expression

Latent Class	Response Parameters
	α0 = onset intensity	αx = rise time (in secs)	αy = peak intensity
High Fear Intensity	0.15	17.19	1.31
Medium Fear Intensity	0.25	12.71	0.93
Low Fear Intensity	0.11	14.03	0.45

3.5. Covariates of class membership

Class-specific means for study covariates are presented in Table 5. Table 6 shows the likelihood of class membership based on child and contextual factors. Using the low fear intensity group as the reference class, the odds of being classified into the high fear intensity group was 0.73 times lower when children exhibited better emotion knowledge and 0.05 times lower when foster parents were more responsive. The odds of being classified into the medium fear intensity group, when compared with the reference low fear intensity group, was 1.14 times higher for children who exhibited more behavior problems. Finally, the odds of being classified into the high fear intensity group over the medium fear intensity group was 0.68 times lower for children with better emotion knowledge, 0.03 times lower for children with more responsive foster parents, and 0.98 times lower for foster parents who modeled appropriate boundaries and behaviors.

Table 5.

Class-Specific Means and Standard Deviations for Study Covariates

	High Fear Intensity		Medium Fear Intensity		Low Fear Intensity
	M	SD	M	SD	M	SD
Child emotion knowledge	42.2	9.34	54.57	4.79	50.48	11.05
Child aggressive behaviors	13.0	7.71	12.71	6.16	9.87	7.97
Child attention problems	5.60	6.11	3.57	2.99	3.37	3.18
Length of time in current placement (months)	12.2	8.20	19.71	21.70	23.35	17.96
Placement transitions	3.0	1.87	4.13	2.42	2.56	1.42
Caregiver responsiveness	5.0	0.71	6.29	0.95	6.0	0.98
Caregiver modeling	3.60	0.55	4.43	0.54	4.00	0.78

Table 6.

Likelihood of Class Membership based on Child and Contextual Covariates

	High compared with Low (reference) fear intensity		Medium compared with Low (reference) fear intensity		High compared with Medium (reference) fear intensity
	OR	95% CI	OR	95% CI	OR	95% CI
Child emotion knowledge	0.73*	0.53, 0.99	1.07	0.96, 1.19	0.68*	0.47, 0.99
Child aggressive behaviors	1.11	0.99, 1.25	1.14*	1.03, 1.25	0.98	0.83, 1.15
Child attention problems	1.71	0.80, 3.63	0.701†	0.47, 1.04	2.44	0.98, 6.11
Length of time in current placement (months)	0.88	0.77, 1.01	0.94	0.84, 1.06	0.94	0.77, 1.14
Placement transitions	1.59	0.62, 4.05	1.33	0.80, 2.22	1.19	0.43, 3.33
Caregiver responsiveness	0.05***	0.00, 0.61	1.44	0.59, 3.53	0.03***	0.00, 0.83
Caregiver modeling	0.12	0.01, 1.80	3.65	0.75, 17.78	0.98***	0.00, 0.658

^*p < .05

^**p < .01

^***p < .001

^†p < 0.10

4. Discussion

Within the current exploratory study, we employed novel approaches to understand heterogeneity in the dynamic patterns of fear arousal and modulation in young children with a history of maltreatment. Based on a hypothesized rise and fall pattern in the moment to moment regulation of fear, we found promising evidence that young foster children could be classified into 3 distinct trajectory classes. Through these preliminary findings, meaningful comparisons between 3 trajectory classes were possible with the use of a re-parameterized version of the quadratic model. Specifically, preschoolers’ fear responses were categorized in the following 3 classes: low fear intensity (lowest onset and peak intensity); medium fear intensity (highest onset intensity and fastest to reach peak intensity); and high fear intensity (highest peak intensity but longest to reach peak). These exploratory findings lend promising support to the notion that regulatory processes are diverse, particularly when heterogeneity exists in how children’s regulation of fear unfolded over time as understood through the response parameters of emotional dynamics (Thompson, 1990; 1994). Further, findings from our exploratory study, extends the work of Maughan and Cicchetti (2002), who documented the presence of dysregulation in young maltreated children.

Additionally, inclusion of child and family-related covariates were associated with these patterns of dynamic response parameters. Specifically, children’s emotion knowledge, aggressive behaviors, attention problems, caregiver responsiveness, and caregiver modeling predicted the likelihood that children were classified into specific fear regulation patterns differentiated by the response parameters of onset intensity, peak intensity, and rise time. These findings are consistent with evidence that documents the link between emotion regulation in young children and these types of child-specific and contextual factors (AUTHOR 3B; Lunkenheimer et al., 2017; Milojevich & Haskett, 2018; Rogosch et al., 2010).

4.1. Heterogeneity in fear regulation, emotional security, and associated risk

Framed by EST-R, evidence for children’s goal of preserving emotional security can be inferred from emotional reactivity, which is characterized by intense and prolonged dysregulated reactions to distress (Davies, Cummings, & Winter, 2004; Davies, & Martin, 2013). Patterns of reactivity typically emerge as an adaptive solution to overcoming distress through the operation of the SDS, resulting in specific profiles of security. Unfortunately, repeated exposure to high levels of unresponsive, unstable, and hostile family contexts have led to difficulties preserving emotional security (Coe, Davies, & Sturge-Apple, 2018; Davies, Winter, & Cicchetti, 2006). For children with a history of maltreatment, examining these profiles of emotional security is important in understanding developmental sequelae, particularly given the variability of outcomes when they are placed in foster care (Lawrence, Carlson, & Egeland, 2006). Used as indicators of emotional reactivity, the fear regulation patterns in this study provided preliminary evidence that young foster children can be classified into specific SDS profiles of security.

First, children who exhibited high fear intensity included patterns of child and context factors that were consistent with the dominant profile of security. Children in this class were more aggressive and impulsive as evidenced by having the highest mean score on measures of aggressive behaviors and attention problems. Children also exhibited the least amount of emotion understanding and had caregivers who were the least responsive and scored the lowest on modeling appropriate boundaries. Although placement instability did not predict class membership, descriptive examination of the class-specific means showed that these children lived in a context with high levels of instability with an average of 3 foster placements (Webster et al., 2000) and by being in their current placement the shortest amount of time.

Second, children who exhibited medium fear intensity included patterns of child and contextual factors that were indicative of the mobilizing profile. Although children in this class showed the highest level of emotion understanding, they still exhibited high levels of aggressive behaviors and attention problems. Consistent with theory, these children had foster mothers who were responsive and showed an appropriate amount of modeling in the home.

Finally, for children who exhibited low fear intensity, interpretation of a corresponding SDS profile of security was more challenging. Relative to the other groups, children in this group had the lowest mean score on both aggressive behaviors and attention problems, indicating lower levels of impulsivity. This finding corresponds with child factors related to either the secure or demobilizing security profile. Consideration of the family context may indicate that these patterns are consistent with the secure profile given the high levels of responsiveness and modeling behaviors of the foster mothers. This is consistent with studies that show the mitigating effects of placement stability and positive caregiving relationships on children’s negative outcomes due to early experiences of maltreatment (Dozier et al., 2006; Fisher, Van Ryzin, & Gunnar, 2011). Alternatively, children’s reactive pattern within this group may correspond to the demobilizing profile. Given that all the children in the sample had substantiated cases of maltreatment that warranted removal from their birth families, children in this group may be exhibiting a coping strategy known as compulsive compliance wherein negative behaviors are usually suppressed or inhibited in moments of distress or maternal directives (Crittenden & DiLalla, 1988; Koenig, Cicchetti, & Rogosch, 2000). Earlier placement into foster care has been shown to mitigate the effects of early adversity (Wade et al., 2018). Unfortunately, age of first placement for children in our study was 24 months or older, which suggests that prolonged exposure to adverse environments may have long-lasting effects on the formation of children’s emotional demobilizing security pattern. The positive caregiving environment may have served as a protective factor and explain increased stability despite this reactive profile of fear regulation.

4.2. Implications for practice

SDS profiles of security have been associated with unique patterns of adjustment in children that include cascading effects on the development of behavior problems (Davies et al., 2016) and later social challenges (Brooker et al., 2016; Buss et al., 2013). Given the promising evidence of heterogeneity in fear response patterns from our exploratory study, interventions should consider individual differences in young foster children’s emotional behaviors, particularly in the reactive responses that correspond to each SDS profile of security using an EST-R framework.

Within the context of child-focused interventions, key response parameters of emotional dynamics should be considered when teaching specific regulatory strategies. For example, lowering onset intensity prior to a fear-eliciting activity may require interventions targeting antecedent-focused emotion regulation strategies, which include attentional re-deployment or cognitive re-appraisals (Gross & Thompson, 2007). In situations where full-blown emotional responses are in effect, response-focused emotion regulation strategies should consider rise time and peak intensity. That is, faster rise time to peak should include strategies that quickly diffuse or modulate an intense reaction whereas a slower rise time to peak may provide opportunities to test the effectiveness of regulatory strategies to decrease an intense emotion. If demobilizing patters are seen where an expected response is not seen, regulatory strategies may need to focus on the up-regulation of a necessary mounted response. For example, in the face of a fear-eliciting activity, development of a healthy regulatory strategy to respond may be needed in order to achieve a goal.

Within the context of parent-child interventions, caregiver modeling of appropriate boundaries and caregiver responsiveness are important intervention targets to consider within the context of specific security profiles. This is particularly important to consider in foster families where stability and consistency of caregiving behaviors are important in organizing young children’s reactive responses (Davies et al., 2016; Davies et al., 2006). For early intervention work, parenting interventions that work with foster parents to improve parent-child relationships are important, such as Child Parent Psychotherapy (Lieberman, Van Horn, & Ippen, 2005) and Multidimensional Treatment Foster Care (Fisher, Kim, & Pears, 2009). Specifically, helping foster parents recognize and work with young children’s specific trajectory patterns is necessary in teaching how to implement regulatory strategies through modeling and teaching.

4.3. Limitations and future directions

First, given the small sample size and lack of a comparison non-maltreated group, results from our study may not be generalizable beyond the characteristics of our sample and caution must be exercised in terms of inferential interpretation of the classes. Because of the exploratory nature of our study, follow-up confirmatory analyses should be conducted in order to examine the extent to which similar classes replicate with larger samples. Despite this limitation, however, post-hoc model convergence and an assessment of model adequacy (e.g., entropy and LRT values) indicated ample power to detect class separation based on recommendations by Masyn (2013). These results are similar to other studies that included smaller sample sizes using linear and non-linear growth models (e.g., MacNeill et al., 2018) and within a growth mixture model framework (e.g., Ram & Grimm, 2009).

Second, caution should be exercised in the interpretation of the fear regulation classes as truly representing each corresponding SDS security profile. The lack of temporally-appropriate distal outcomes that trace the developmental sequelae of each hypothesized security profile is needed to further validate the classes derived in our study. However, the inclusion of theoretically-driven covariates still provided ample evidence to show how each fear regulation class might correspond with each specific security profile.

Finally, interpreting the dynamic changes in response features of fear as influenced solely by each security profile should be done with restraint. Regulatory strategies are often employed in the modulation of emotions and given that specific strategies were not explicitly measured in our model, it is difficult to disentangle its relative influence on how onset intensity, rise time, or peak intensity may have shifted. Even without the inclusion of regulatory strategies, however, evidence for modulation of fear could be inferred based on the appropriateness of the theoretically-informed and empirically-derived quadratic model.

Future work should address the limitations highlighted above by increasing the sample size and including a comparable non-maltreated sample. Inclusion of a comparison group provides an opportunity to test the viability of a 4-class model, which is in line with the 4 SDS profiles of security. A comparison group would also allow a more careful examination of the role of placement instability, which contributed to the experiences of the children in this study but did not predict class membership. Future work should also include a second wave of data collection in order to use these trajectory classes to predict a distal outcome based on the hypothesized sequelae of development for each SDS profile (Davies et al., 2016). Inclusion of a distal outcome to validate each class is typically a third step in typical mixture modeling analyses. Finally, future studies should include explicit emotion regulation strategies as covariates in the model to examine its effect on the pattern of fear regulation exhibited by young foster children.

5. Conclusion

Despite the limitations outlined earlier, findings from our study provided evidence for heterogeneity in the dynamic expression and modulation of fear in a sample of young foster children. Specifically, we were able to identify 3 distinct classes of preschoolers’ fear responses: low fear intensity; medium fear intensity; and high fear intensity. We also found that child and contextual factors predicted class membership, particularly children’s emotion knowledge, children’s behavior problems, and foster parenting (i.e., responsivity and modeling of behaviors).

These findings contribute to research in the field by providing a novel approach to modeling these dynamic regulatory processes that allow for a more person- and process-oriented perspective, highlighting individual differences in young children’s emotional responding to threat while considering contextual influences. Moreover, this study fills a large gap in the child welfare field regarding the social-emotional outcomes of young children exposed to maltreatment and foster care, a research landscape that tends to focus on their behavior problems. From a practice perspective, this evidence supports the implementation of child-directed interventions targeting the emotion regulation of young children in the foster care system, as well as parenting interventions to facilitate the emotion socialization skills of foster parents.

Highlights:

• Early adversity result in heterogeneity in emotion regulation patterns

• Novel methods capture heterogeneity in the dynamic rise and fall of fear regulation

• Child and caregiver factors predict latent class membership