Harsh Parenting, Parasympathetic Activity, and Development of Delinquency and Substance Use

Abstract

Stress response systems are thought to play an important role in the development of psychopathology. Additionally, family stress may have a significant influence on the development of stress response systems. One potential avenue of change is through alterations to thresholds for the activation of stress responses: Decreased threshold for responding may mark increased stress sensitivity. Our first aim was to evaluate the interaction between thresholds for parasympathetic nervous system (PNS) responding, operationalized as resting respiratory sinus arrhythmia (RSA), and harsh parenting in the prediction of development of delinquency and adolescent substance use (resting RSA as a biomarker of risk). The second aim was to evaluate if resting RSA changes over time as a function of harsh parenting and stress reactivity indexed by RSA withdrawal (altered threshold for stress responding). Our third aim was to evaluate the moderating role of sex in these relations.

We used longitudinal data from 251 children ages 8 to 16. Mother-reports of child delinquency and RSA were acquired at all ages. Adolescents self-reported substance use at age 16. Family stress was assessed with child-reported harsh parenting.

Controlling for marital conflict and change over time in harsh parenting, lower resting RSA predicted increases in delinquency and increased likelihood of drug use in contexts of harsh parenting, especially for boys. Harsh parenting was associated with declining resting RSA for children who exhibited greater RSA withdrawal to stress. Findings support resting PNS activity as a moderator of developmental risk that can be altered over time.

Stress sensitivity can be a key mechanism in the development of psychopathology. It can be conceptualized in multiple ways, including emotional or physiological hyperarousal in the context of ongoing stress (Bale, 2006) as well as increased susceptibility to subsequent negative outcomes (e.g., problem behavior) given exposure to stress (e.g., harsh parenting) (Boyce & Ellis, 2005). In the present paper, we consider whether low resting respiratory sinus arrhythmia (RSA) serves as a biomarker of stress sensitivity, by examining whether the prospective association between harsh parenting and problem behaviors is stronger among children with lower resting RSA compared to children with higher resting RSA (Aim 1). Next, we consider how this biomarker of stress sensitivity (low resting RSA) emerges developmentally, by examining whether harsh parenting particularly predicts declining resting RSA among children who exhibit relatively greater RSA withdrawal) under stressful conditions (Aim 2).

Low Resting RSA as Biomarker of Stress Sensitivity (Aim 1)

RSA is a measure of parasympathetic nervous system (PNS) regulation of heart rate. The PNS serves as a “brake” (via the vagus nerve) that decelerates heart rate, producing higher RSA, and facilitates emotional regulation and social engagement under normal circumstances (Berntson et al., 1997; Porges, 2007). The regulatory, or calming, function of the PNS is explained by the inhibitory influence of the vagus nerve on the cardiac pacemaker, and the social engagement function stems from the integration of neural networks that regulate the vagus nerve and muscles involved in communication (i.e., speaking, hearing, orienting, and facial expressions). Under threatening or challenging conditions, the vagal brake can be withdrawn (i.e., vagal withdrawal or reduction in RSA), yielding an incremental and efficient increase in arousal and metabolic output that may facilitate engagement with environmental demands.

One advantage of high resting RSA is that it provides a greater dynamic range in which RSA withdrawal may occur without yielding extremely high levels of cardiovascular arousal, whereas low resting RSA may create a decreased threshold for stress responding and contribute to hyper-arousal (Beauchaine, 2001; Porges, 2007). Indeed, lower resting RSA has been linked with elevated externalizing behavior problems (Beauchaine, 2001; Beauchaine, Gatzke-Kopp, & Mead, 2007) as well as internalizing symptoms (Dietrich et al., 2007; Forbes, Fox, Cohn, Galles, & Kovacs, 2006). Furthermore, low resting RSA in conjunction with greater stress-linked RSA withdrawal predicts heightened internalizing symptoms in late childhood (Hinnant & El-Sheikh, 2009). Thus, children with low resting RSA may exhibit stress sensitivity, or hyperarousal in the context of stress, when they experience elevated or even normal (and otherwise adaptive) levels of stress-linked RSA withdrawal.

Low resting RSA also may be characterized as a marker of stress sensitivity to the extent that it increases children’s susceptibility to negative outcomes in the context of stress. Harsh parenting is a salient stress experience in childhood, referring to coercive behaviors and negative emotional expressions that parents direct toward children, including verbal and physical aggression. Numerous longitudinal studies have shown that harsh parenting predicts problem behaviors in childhood and adolescence (e.g., Lansford et al., 2011; Gershoff, 2002). Noncompliant and aggressive behaviors are common problem behaviors that may stem from harsh parenting in childhood; in adolescence, problem behaviors commonly take the form of delinquency and substance use (Dishion & Patterson, 2006). Consistent with problem behavior theory (Jessor & Jessor, 1977), multiple problems behaviors (e.g., lying, stealing, substance use) tend to co-occur and are remarkably costly in terms of human suffering and societal expenditures (Angold, Costello, & Erklani, 1999; Foster, Jones, & the Conduct Problems Prevention Research Group, 2006; Mayes & Suchman, 2006).

Well-established mechanisms connecting harsh parenting with problem behaviors include social learning and negative reinforcement of coercive behaviors (Patterson, Reid, & Dishion, 1992) and acquisition of hostile social-cognitive schemas and social information processing patterns (Dodge, Bates, & Pettit, 1990). Physiological underpinnings of problem behaviors have been identified (e.g., Raine, 2002) and featured in developmental models of problem behaviors (e.g., Beauchaine et al., 2007; Dodge & Pettit, 2003). However, research on the interactive contribution of harsh parenting and autonomic activity to problem behaviors in childhood and adolescence is limited.

In a cross-sectional study of middle childhood, the association between parent-child aggression and adjustment problems was stronger among children with lower resting RSA or less RSA withdrawal (decline from resting to challenge conditions) compared to children with higher resting RSA or greater RSA withdrawal (Whitson & El-Sheikh, 2003). Although no other studies have examined interactions between harsh parenting and resting RSA to our knowledge, several studies have examined interactions between other forms of family stress and RSA. For example, associations linking interparental conflict and parental problem drinking with externalizing behavior problems are stronger among children with lower resting RSA compared to children with higher resting RSA (El-Sheikh, 2005a; El-Sheikh, Harger, & Whitson, 2001; El-Sheikh, Hinnant, & Erath, 2011; Katz & Gottman, 1995, 1997).

To our knowledge, no published longitudinal studies have examined whether the vulnerability function of lower resting RSA (or protective function of higher resting RSA) extends to harsh parenting as the source of environmental adversity or to problem behaviors in adolescence as the outcome variable. In the present study, we examined whether lower resting RSA operates as a marker of stress sensitivity, such that the prospective association between harsh parenting in childhood and growth in problem behaviors from middle childhood to middle adolescence is stronger among children with lower resting RSA compared to children with higher resting RSA.

Stress Sensitivity (Low RSA) as a Developmental Phenomenon (Aim 2)

If low RSA is a marker of stress sensitivity (Aim 1), it is important to understand how RSA develops over time and the contexts that contribute to its altered functionality (Aim 2). RSA is somewhat heritable (Wang et al., 2009) and moderately stable in childhood and early adolescence (Calkins & Keane, 2004; El-Sheikh, 2005b; Hinnant, Elmore-Staton, & El-Sheikh, 2011), yet susceptible to environmental influence. In particular, emerging research provides evidence that family stress may shape RSA. For example, Rigterink, Katz, and Hessler (2010) reported that children exposed to domestic violence exhibited a smaller increase in RSA from ages 4.5 to 9 years, compared to non-exposed children. Similarly, children of parents with a mood disorder exhibited smaller increases in RSA during late childhood, compared to children of parents without a mood disorder (Gentzler, Rottenberg, Kovacs, George, & Morey, 2012). Several cross-sectional studies have also shown that poor parent-child relationships or marital conflict are associated with lower resting RSA or less RSA withdrawal in the context of stress among infants and young children (for a review, see Propper & Holochwost, 2013), although at least one other cross-sectional study yielded contrasting results in which higher marital conflict was associated with higher resting RSA among young children (Davies, Sturge-Apple, Cicchetti, Manning, & Zale, 2009).

As noted with respect to Aim 1 of the current study, vagal tone (operationalized via resting RSA) may be conceptualized as an index of stress sensitivity based on its protective or vulnerability function in the context of environmental stress and based on the threshold it sets for stress responding. As Aim 2, the current study also examined whether resting RSA may erode (i.e., decline) over time under pressure of harsh parenting. We further considered whether harsh parenting may particularly predict declining resting RSA among children who exhibit greater RSA withdrawal to stress. This hypothesis is guided by relevant work on marital conflict and problem behaviors in childhood. In particular, El-Sheikh and Hinnant (2011) recorded decreasing resting RSA from ages 8 to 11 among boys with greater RSA withdrawal to a frustrating laboratory task in conjunction with high or increasing exposure to marital conflict.

This prior study and our current hypothesis is consistent with the allostatic load framework (McEwen & Stellar, 1993), which contends that recurring activation of physiological stress responses (e.g., RSA withdrawal) in the context of intense or chronic environmental stress (e.g., harsh parenting) may incur “wear and tear” on these systems and contribute to their dysregulation, or lowered threshold for stress responding, over time (e.g., low resting RSA), potentially yielding increased sensitivity to stress. Homeostasis refers to balance and stability in the state of physiological systems, whereas allostasis refers to re-establishing homeostasis under conditions of change or stress. Intense or chronic environmental stress, especially if accompanied by physiological reactivity, may create allostatic load and eventual damage to physiological systems over time (McEwen & Stellar, 1993). The present study examined whether harsh parenting predicts declining RSA from middle childhood through middle adolescence among children with greater RSA withdrawal to stress, controlling for changes in harsh parenting from middle childhood to middle adolescence.

Sex as a Moderator of Developmental Phenomenon (Aim 3)

Some studies provide evidence for sex differences in associations among family stress, RSA, and problem behaviors. For example, boys with clinical levels of aggressive behavior exhibited lower resting RSA compared to boys with lower levels of aggression while no differences were found among girls (Beauchaine, Hong, & Marsh, 2008). Similar sex differences have been found in a community sample in which lower resting RSA was related to higher problem behaviors among boys but not among girls (Calkins & Dedmon, 2000). El-Sheikh et al. (2011) reported that destructive marital conflict predicted increases in problem behaviors over time, especially among boys with lower resting RSA and weaker RSA withdrawal. In addition, El-Sheikh and Hinnant (2011) found that boys with lower resting RSA exhibited increases in problem behaviors from ages 8 to 11, and, as noted, boys with greater RSA withdrawal who were exposed to elevated marital conflict exhibited declining resting RSA over time, whereas results were less consistent among girls. These findings, while not conclusive, support the notion that: (1) sex may be an additional moderator of biosocial interactions that helps to differentially predict the development of psychopathology symptoms in externalizing and internalizing domains (Aim 1); and (2) sex as a moderator may help to explain individual differences in the development of stress sensitivity and allostatic calibrations of stress response systems (Aim 2). Thus, the present study explored sex differences in associations linking harsh parenting with resting RSA and problem behaviors from childhood to adolescence (Aim 3).

The Current Study

In summary, the current study examined: (1) whether resting RSA (Time 1; age 8) moderates associations between child-reported harsh parenting in childhood (Time 1; age 8) with mother-reported delinquency from childhood through adolescence (growth from Time 1 – Time 4; ages 8 – 16) and adolescent-reported substance use (Time 4; age 16); (2) whether RSA withdrawal to a well-validated laboratory stress task (star-tracing task; T1) moderates the association between harsh parenting and growth in resting RSA from childhood to adolescence (T1 – T4); and (3) whether sex moderates associations among harsh parenting, resting RSA, RSA withdrawal, and problem behaviors. All associations were tested with control for potential demographic confounds (i.e., age, ethnicity, income) as well as marital conflict and changes in harsh parenting over time. We anticipated that lower resting RSA would exacerbate the association between harsh parenting and problem behaviors and would stem from harsh parenting in conjunction with greater RSA withdrawal to stress. We also anticipated that these associations would be stronger among boys than girls based on existing empirical evidence. These hypotheses were tested with a large and diverse sample who participated in a multi-wave, multi-method study that involved intensive assessments of family stress and physiological regulation in childhood as well as assessments of highly significant mental and behavioral health outcomes in adolescence, including delinquency and substance use.

Methods

Participants

Children and their parents participated in four waves of data collection from childhood to adolescence as part of the Family Stress and Youth Development: Bioregulatory Effects project. There were one-year intervals between the first, second, and third assessments and a five-year interval between the third and fourth assessments. Data collection occurred in 2005 (age 8), 2006 (age 9), 2007 (age 10) and 2012–2013 (age 16). School-aged children and their parents were recruited from elementary schools in the Southeastern United States. At age 8, children in second- or third-grade from two parent homes (cohabitating for at least two years) were eligible for participation. Further, children diagnosed with attention deficit hyperactivity disorder, a developmental or learning disability, or a chronic illness were not eligible to participate due to potential confounds.

At the first assessment, participants included 251 children (122 boys, 129 girls; M age = 8.23 years, SD = .72) and their parents. A majority of children (74%) lived with both biological parents and 26% lived with a parent and stepparent (mostly the mother and stepfather/partner). Of children who participated at age 8, 86% (N = 217) participated at the second assessment (106 boys, 111 girls; M age = 9.31 years, SD = .79), and of those who participated at age 9, 84% (N = 194) participated at the third assessment (92 boys, 102 girls, M age = 10.28 years, SD = .99). Lastly, 83% (N = 199) of children who participated at any of the three earlier assessments participated at the fourth assessment at age 16 (93 boys, 106 girls, M age = 15.79 years, SD = .81). To increase power, an additional 53 families were recruited to participate at age 16 (25 boys, 28 girls, M age = 15.26 years, SD = .90), though data from these families have limited influence in the predictive models because all of the main predictors were collected at age 8. Reasons for attrition included a lack of interest in participating and geographic relocation. There were no differences on study variables among participants who did and did not participate in the second and third assessments. Annual family income was scored on a 6-point scale (1 = less than $10,000 to 6 = more than $75,000). Participants who did not participate in the fourth assessment had lower annual family income at age 8 compared to those who participated at age 16 (M = 4.03, SD = 1.46; t = −2.98, p < .01). Further, those who did not participate at age 16 had less RSA withdrawal (M =.00, SD = .11) at age 8 and higher resting RSA (M = .20, SD = .11) at age 10, compared to those who participated at age 16 (age 8 RSA withdrawal: M = −.03, SD = .06, t = 2.16, p < .05; age 10 resting RSA: M = .16, SD = .08, t = 2.16, p < .05). Additionally, children participating for the first time at the first versus the fourth assessments did not differ on demographic or primary study variables. All available data were used in analyses. Proportions of missing data are given under each measure below.

The sample consisted of approximately 64% to 66% European American and 34% to 36% African American children across study waves, which is representative of the community from which participants were recruited. The sample was diverse in relation to family income. Across all assessments, the percent of families reporting the following ranges of annual income was: 13% to 17% for income < $20,000; 31% to 47% for $20,000 – $50,000; and 21% to 26% for $50,000 – $75,000. Finally, families reporting income > $75,000 ranged between 16% and 30%. Data on ethnicity was available for all participants and data on family income at age 8 was available for 95% of participants.

Procedure

This study is part of a larger, longitudinal investigation and only pertinent procedures are discussed. The study was approved by the University’s Institutional Review Board. Consent and assent for participation were obtained from parents and children at each study wave. Data were collected at a university laboratory. At T4, some participants completed questionnaires at home versus in the laboratory. Children and parents visited during the same session and completed questionnaires in separate rooms. All procedures and measures are identical across the 4 study waves unless otherwise specified.

Children’s RSA was collected and analyzed following standard guidelines (Berntson et al., 1997). Electrodes were placed on the child’s chest using a standard or modified lead-II configuration to measure electrocardiogram (ECG) activity and respiration. There was a 3-to 6-minute adaptation period, during which children were asked to sit quietly and relax. This was followed immediately by a 3-minute baseline/resting measurement during which children were sitting quietly and then by a 3-minute star-tracing task in which children traced an outline of a star on a sheet of paper while looking into a mirror to guide their movements (Mirror Tracer; Lafayette Instrument Co., Lafayette, IN). This frustrating task consistently elicits significant stress responses in multiple physiological systems, including RSA withdrawal (e.g., El-Sheikh et al., 2011).

For the first, second, and third assessments, RSA was examined with hardware and software from the James Long Company (Caroga Lake, NY). In addition to the aforementioned electrodes, a pneumatic bellows was firmly placed around the chest to measure respiratory changes. The ECG signal was digitized at a sampling rate of 1,000 readings per second using bandpass filtering with half power cutoff frequencies of .1 and 1,000 Hz and a gain of 500. The Interbeat Interval (IBI) Analysis System from the James Long Company (Caroga Lake, NY) was used to process the ECG signal. An automated algorithm identified R-waves. In the rare case that it was needed, an interactive graphical program was used for manual correction of misidentified R-waves. R-wave times were then converted to IBIs, resampled into equal time intervals of 125 ms, prorated and stored for computation of RSA. RSA was calculated using the peak-to-valley method, which is an acceptable approach for quantifying RSA (Bernston et al., 1997). Units of measure for RSA from ages 8 to 10 were in seconds.

At the fourth assessment, adolescents’ RSA was collected and analyzed with equipment and software from MindWare Technologies Ltd (Gahanna, OH). Data were collected via the MW1000A acquisition system (MindWare Technologies). A MindWare BioNEx 8-slot chassis was used to collect ECG data. Cardiovascular responses were recorded with the ECG activity amplifier module. Respiration was recorded through four disposable Ag/AgCl ECG electrodes; two electrodes were placed about 3 cm apart on the participant’s chest, and two electrodes were positioned about 3 cm apart vertically on the mid and lower spine, and was calculated through spectral analysis of thoracic impedance (Z₀; Ernst, Litvack, Lozano, Cacioppo, & Berntson, 1999). Physiological data were scored in 1-minute intervals with MindWare software (Heart Rate Variability Version 3.0.21). The cardiovascular data was inspected for artifacts and missing R peaks on the basis of improbable interbeat intervals. Missing or misplaced R peaks were inserted manually. Units of measure for RSA at the fourth assessment were converted from milliseconds to seconds to match the units of measure for RSA from the first to third assessments. Importantly, Grossman and colleagues have found these two methods of calculating RSA (i.e., peak-to-valley and spectral approaches) to be very similar, with within-individual correlations of the two methods to be about .96 (Grossman, van Beek, & Wientjes, 1990). In the current study we found the correlation between RSA scores in the original and rescaled metric from msec to sec to be .99. Correlations between RSA scores at each assessment are given in Table 1. For the first three assessments correlations ranged from between .56 to .65 while correlations between the first three assessments and the fourth assessment were somewhat lower, ranging from .35 to .43. We note, however, that the lag between the third and fourth assessments was 5 years, rather than the one year lag between the earlier assessments. Overall reliability (Cronbach’s alpha) for the four repeated measures of resting RSA from ages 8 to 16 was .80.

Table 1.

Descriptive Statistics and Correlations among Study Variables

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20
1. Sex	-
2. Race	−.04	-
3. Age	−.20**	.11	-
4. Income	.04	.35**	−.02	-
5. Marital conflict	−.14*	−.05	.03	−.09	-
6. Harsh parenting	−.22**	−.09	−.06	.03	.41**	-
7. Harsh parenting Δ	.19**	.02	.04	−.07	−.25**	−.83**	-
8. RSA withdrawal	−.04	.14*	−.01	−.02	−.05	−.10	.07	-
9. Age 8 Resting RSA	−.02	−.28**	.03	−.17**	−.02	−.03	.06	−.41**	-
10. Age 9 Resting RSA	.04	−.16*	−.04	−.04	−.04	−.03	.04	−.04	.63**	-
11. Age 10 Resting RSA	−.05	−.12	−.03	−.16*	−.02	.05	−.02	−.05	.56**	.65**	-
12. Age 16 Resting RSA	.10	−.17*	−.11	.02	−.16*	.04	−.06	−.05	.35**	.45**	.43**	-
13. Age 8 Delinquency	.05	.01	.05	−.16*	.06	.18**	−.14	−.02	.02	−.09	−.02	−.11	-
14. Age 9 Delinquency	−.10	.06	.15*	−.13	.08	.16*	−.13	.03	−.04	−.11	.01	−.04	.73**	-
15. Age 10 Delinquency	−.01	.08	.05	−.17*	.03	.13	−.04	.07	−.02	−.01	.10	−.01	.66**	.80**	-
16. Age 16 Delinquency	−.02	−.08	.02	−.23**	.03	.17*	−.21*	−.01	.12	.07	.06	−.03	.43**	.53**	.51**	-
17. Tobacco use	−.11	.08	.21**	−.08	.03	−.05	.06	.04	.18*	.17*	.06	.09	.04	.09	.06	.29**	-
18. Alcohol use	−.04	−.05	.15*	−.11	.02	−.08	.07	−.01	−.02	.07	.01	−.02	.06	.11	.05	.20**	.50**	-
19. Marijuana use	−.14	−.02	.22**	−.20**	−.01	−.05	.03	−.02	.19*	.16*	.08	.07	.13	.18*	.16*	.33**	.63**	.45**	-
20. Other drug use	−.03	.12	.06	−.07	.11	.00	−.05	−.01	.04	.07	.06	.03	−.10	−.12	−.14	.03	.35**	.32**	.37**	-
Mean (SD)	-	-	98.71 (8.64)	3.87 (1.46)	2.09 (3.30)	5.47 (4.90)	−2.56 (4.73)	−.02 (.06)	.15 (.08)	.16 (.08)	.16 (.09)	.17 (.03)	1.88 (1.00)	1.90 (1.15)	1.74 (1.17)	1.86 (1.42)	.20 (.40)	.44 (.50)	.19 (.39)	.09 (.28)

Note. Sex coded 0 = boys, 1 = girls; race coded 0 = African American, 1 = European American; substance use coded 0 = has not engaged in use, 1 = has engaged in use; age assessed in months at the first assessment. Harsh parenting Δ refers to change in harsh parenting: Age 16 scores – age 8 scores. Delinquency scores are square root transformations to correct for skew. RSA = respiratory sinus arrhythmia.

^*p < .05.

^**p < .01.

Measures

Harsh parenting (age 8)

Children completed the Parent-Child Conflict Tactics Scale (CTSPC; Straus, 1999), which is widely used to assess harsh parenting (Yodanis, Hill & Straus, 2001). Psychometric properties of the CTSPC have been well-established (Straus, 1999). Children reported on their mothers’ and fathers’ frequency of psychological/verbal (e.g., shouted, yelled, or screamed; five items) and physical (e.g., spanked you on the bottom with bare hand; slapped you on the face or head or ears; nine items) aggression directed toward the child in the past year. Items were rated on a 7-point scale; a score of 0 = did not happen in the past year; 1 = once; 2 = 2 times; 3 = 3 – 5 times; 4 = 6 – 10 times; 5 = 11 – 20 times; 6 = > 20 times. Mothers’ and fathers’ psychological and physical aggression were first averaged in order to assess reliability of separate parental psychological and physical aggression scales. Internal consistency for psychological/verbal and physical aggression were high (αs = .79 and .83, respectively). Approximately 89% of youths reported experiencing physical aggression from at least one parent, and a wide range of verbal aggression was reported (0 to 28.50). Reports of verbal and physical aggression were highly correlated (r = .62) and were averaged to create a harsh parenting score. Data on child-reported harsh parenting was available for 99% of participants at age 8.

RSA (ages 8 – 16)

Resting RSA was obtained at each time point and RSA during the star-tracing task was obtained at age 8 to compute RSA withdrawal scores. From ages 8 to 10, RSA was computed as the difference in IBI readings from inspiration to expiration onset. At age 16, RSA was calculated as the natural log of the high frequency power (.15 – 40 Hz). Additionally, at age 8, RSA withdrawal to the star-tracing task was computed as a difference score (RSA during star-tracing minus resting RSA). Lower RSA withdrawal scores indicate greater RSA withdrawal from resting to task conditions. Data for RSA withdrawal to stress at age 8 was available for 94% of participants. Data for repeated measures of resting RSA was available for 96%, 77%, 71%, and 72% of participants at ages 8, 9, 10, and 16, respectively. Of the original 251 participants, resting RSA data at age 16 was available for 67% of adolescents.

Delinquency (ages 8 – 16)

Mothers reported on children’s delinquent behaviors using the Personality Inventory for Children-II (PIC2; Lachar & Gruber, 2001). The PIC2 has demonstrated discriminant and construct validity, interrater reliability, and test-retest reliability (Lachar & Gruber, 2001; Wirt, Lachar, Klinedinst, & Seat, 1990). Youths’ symptoms on the Delinquency scale consisted of 47 items that assess antisocial behavior (e.g., “My child has been in trouble with the police”), dyscontrol (e.g., “My child often looks for a fight or argument”), and noncompliance (e.g., “My child often breaks the rules”). Reliabilities for the delinquency scale raw scores at ages 8, 9, 10, and 16 were .80, .90, .98 and .92, respectively. At age 8, 17 children (6%) had scores in the borderline or clinical range (T ≥ 60). At ages 9 and 10, 14 (6%) and 12 (6%) of children, respectively, had scores in the borderline or clinical range. At age 16, 32 (13%) adolescents had scores in the borderline or clinical range. Although referred youth exhibit significantly higher delinquency scores than do control youth and clinician descriptions of youth in the clinical range match the content of the scale (Lachar & Gruber, 2001), the PIC2 is not a tool for clinical diagnosis (Frick, Barry, & Kamphaus, 2010) and is most appropriately used to indicate relative risk of individuals and need for further evaluation. Data for repeated measures of mother-reported delinquency was available for 96%, 83%, 75%, and 70% of participants at ages 8, 9, 10, and 16, respectively. Of the original 251 participants, mother-reported delinquency scores at age 16 was available for 66% of adolescents.

Substance use (age 16)

Adolescents completed the Centers for Disease Control’s Youth Risk Behaviors Survey (CDC Youth Risk Behavior, 2011; Mrug, Gains, Su, & Windle, 2010). Substance use data for the current report was collected only at age 16. Adolescents reported low frequencies of use of illicit substances which resulted in highly skewed data. Thus, we coded the absence or presence of tobacco, alcohol, marijuana, and other drug use into a binary form suitable for use in logistic regression analyses. “Other drug use” included an array of less common and generally more dangerous drugs such as cocaine and methamphetamine. Adolescents who reported never having used the substance in question received a score of 0 while those who had reported using the substance received a score of 1. Twenty percent of adolescents reported having used tobacco, 44% reported having consumed alcohol, 19% reported having smoked marijuana, and 9% reported having used other drugs. Correlations for substance use ranged from .32 to .63, all p’s < .001. Substance use data at age 16 was available for 76% of participants. Of the original 251 participants, substance use data at age 16 was available for 72% of adolescents.

Control variables

Demographic variables (age 8)

Some variables were associated with the primary constructs in the present study and were controlled in analyses: child sex (boys = 0, girls = 1); age in months; ethnicity (0 = African American; 1 = European American); and annual family income; the latter was scored on a 6-point scale (1 = less than $10,000 to 6 = more than $75,000). Data on sex and ethnicity were available for all participants and family income data were available for 95% of participants.

Marital conflict (age 8)

Because marital and parent-child conflict are associated frequently, and for a more conservative test of our research questions, marital conflict was controlled in analyses. Using the Conflict Tactics Scale (CTS2; Straus, Hamby, Boney-McCoy, & Sugarman, 1996), children reported about mothers’ and fathers’ psychological/verbal (e.g., insulted or swore at him/her; eight items) and physical (e.g., pushed or shoved him/her; 11 items) aggression towards one another in the past year. Items were rated on a 7-point scale (0 = did not happen in the past year to 6 = >20 times). Approximately 25% of children reported marital physical aggression; a wide range of verbal aggression was also endorsed (0 to 27.50). High internal consistency was found for verbal (α = .83) and physical (α = .82) aggression. The two aggression scales were highly correlated (r = .62) and were averaged to create a marital conflict score, which was used as a control variable in analyses. Ninety eight % of marital conflict data were available at age 8.

Change in harsh parenting

To account for the potential effects of change in harsh parenting between ages 8 and 16, we created a difference score to include as a covariate. The averaged physical-verbal CTSPC score at age 16 was subtracted from the average score at age 8, with positive difference scores indicating increased harsh parenting over time. Data on harsh parenting difference scores were available for 73% of participants.

Results

Plan of Analysis

We analyzed the repeated measures data of resting RSA and delinquency with linear mixed models (LMM) in SPSS. LMM is a flexible analytic tool that accommodates correlated repeated measures for assessing within- and between-individual linear and, despite the name, non-linear change over time (West, 2009). In our LMM analyses, intercept and time effects (i.e., repeated measures) were treated as random and individual difference variables (i.e., predictors) were treated as fixed. LMM has other advantages in that it uses Full Information Maximum Likelihood (FIML) estimation, which is recommended for handling missing data (Acock, 2005), and that it does not make the often violated assumption of sphericity that is problematic in repeated measures data. In particular, FIML has been shown to be a superior method for model estimation in the context of data that is missing completely at random or missing at random (Enders & Bandalos, 2001). As the name implies, FIML uses all available data to estimate model parameters along with the sample-specific log likelihood function to increase the accuracy of parameters. LMM does not offer absolute indices of model fit, though when appropriate we report comparative fit indices of Akaike’s Information Criterion (AIC) and Schwarz’s Bayesian Information Criterion (BIC), with smaller values indicating better fit.

We first constructed two unconditional growth models to arrive at unbiased estimates of initial (intercept) levels and change over time (linear and quadratic slopes) in delinquency and resting RSA at ages 8, 9, 10, and 16 (Curran & Willoughby, 2003). Intercepts for delinquency and resting RSA were set at the initial time point (age 8) and time was coded to “count up” over time (linear slope time coding of 0, 1, 2, and 8 and quadratic slope time coding of 0, 1, 4, and 64), which allowed us to assess linear and non-linear patterns of change over time in the dependent variables (Biesanz, Deeb-Sousa, Papadakis, Bollen, & Curran, 2004).

Next, we tested our substantive research questions in two conditional mixed models and a series of logistic regressions. Controlling for individual differences in age, ethnicity, and income as well as exposure to parental marital conflict and change in harsh parenting, in our first question we tested whether child-reported harsh parenting, resting RSA, and sex at age 8 have main and interactive effects on the development of delinquency from ages 8 to 16. In conjunction with these analyses, we evaluated the relations between these variables and substance use at age 16 in four logistic regressions. In our second research question, controlling for the above extraneous variables, we tested whether child-reported harsh parenting, RSA withdrawal to the star tracer task, and sex at age 8 have main and interactive effects on the development of resting RSA from age 8 to 16.

To facilitate interpretation, predictor variables were mean centered (Aiken & West, 1991). Significant interactions were plotted at ± 1 SD and simple slopes were tested to evaluate whether they were significantly different from zero (Aiken & West, 1991; Curran & Willoughby, 2003). Thus, for each significant interaction we reanalyzed the data in order to derive estimates of the main effect on a given outcome at conditional values of interest of the moderator(s) (± 1 SD) to plot figures. In each table of results parameter estimates are presented for those individuals coded 0 on dichotomous variables (i.e., sex is dummy coded 0 for boys and 1 for girls while ethnicity is coded 0 for African American and 1 for European American.

Preliminary Analyses

Descriptive statistics and correlations among study variables are presented in Table 1. Several variables were positively skewed, including harsh parenting, marital conflict, income, and delinquency at each time point. Because skew can result in underestimation of standard errors and increased type 1 error rates (Muthen & Kaplan, 1985), square root transformations were used to normalize these variables and all analyses were conducted with both transformed and raw scores. We found no differences in the results using transformed independent variables. Thus, results are presented with raw scores for independent variables. We did, however, find substantially different results when using square root transformed delinquency scores and present findings with transformed delinquency scores in text and tables. Figures for delinquency results (1A and 1B) have been reverted to their original scales to facilitate interpretation.

Repeated measures of resting RSA and delinquency were correlated within domains, with correlations ranging from .35 to .65 for resting RSA and from .43 to .80 for delinquency. European American children tended to have lower resting RSA than African American children at most time points, and came from families with higher incomes. Sex was not associated with repeated measures of resting RSA or delinquency, but girls reported lower levels of marital conflict and harsh parenting. Income was inconsistently related to lower resting RSA and delinquency. Child-reported marital conflict and harsh parenting were positively correlated. Although marital conflict did not show consistent associations with resting RSA or delinquency, harsh parenting was consistently but weakly associated with higher levels of delinquency. Finally, delinquency at ages 9 and 10 was positively correlated with marijuana use at age 16, and delinquency at age 16 was positively related to concurrent adolescent use of tobacco, alcohol, and marijuana (r’s ranging from .20 to .33, all p’s < .01).

RSA withdrawal to the star-tracer task was associated with higher resting RSA at age 8 (r = −.41, p < .001) but was not associated with resting RSA at later time points, and was not directly related to repeated measures of delinquency or substance use at age 16. RSA withdrawal at age 8 was the most common response to the star-tracing challenge with 68% of children exhibiting some decrease in RSA during the task.

Unconditional Growth Models

Development of delinquency

The average square root transformed delinquency score at age 8 was M = 1.92 (.064), p < .001. The average linear slope was negative and not significantly different from zero, M = −.13 (.08), p = .09, and the average quadratic slope was positive but also not significantly different from zero, M = .03 (.03), p = .24. Each component of the delinquency growth model exhibited significant variability: intercept var = .74 (.10), p < .001; linear slope var = .31 (.15), p = .04; quadratic slope var = .07 (.02), p < .001. Model fit indices were AIC 44.95 and BIC 45.25. Comparing this model to a simpler model including only a linear slope with fit indices of AIC 90.56 and BIC 90.74 indicated that the model with both linear and quadratic slopes was a comparatively better representation of the data. To summarize, there was not significant linear or non-linear change over time in delinquency symptoms, but there was significant individual variability in initial levels and change over time in linear and quadratic slopes, which could be accounted for by between-individual differences.

Development of resting RSA

The average resting RSA at age 8 was M = .154 sec (.005), p < .001. The average linear slope was M = .005 (.006), p = .39, indicating that resting RSA did not show significant mean level change over time. The average quadratic slope was very close to zero, M = −.0006 (.0019), p = .75. Each component of the growth model exhibited significant variability: intercept var = .004 (.0005), p < .001; linear slope var = .0006 (.0001), p < .001; and quadratic slope var = .00001 (.000003), p < .001. Model fit indices were AIC 19.70 and BIC 19.97. Comparing this model to a simpler model including only a linear slope with fit indices of AIC 68.34 and BIC 68.52 indicated that the model with both linear and quadratic slopes was a comparatively better representation of the data. Thus, although there was not significant linear or quadratic change over time in resting RSA, on average, there was significant individual variability in initial levels and change over time in linear and quadratic slopes, which could potentially be accounted for by between-individual differences.

Conditional Growth Models

Predictors of development of delinquency

Results from this model can be found in Table 2. Higher family income was associated with lower levels of delinquency at age 8. Harsh parenting was related to higher levels of delinquency at age 8. No other variables were predictive of delinquency at age 8.

Table 2.

Longitudinal Model Results for Prediction of Delinquency

	Delinquency
	Intercept B (SE)	Linear slope B (SE)	Quadratic slope B (SE)
Estimate	1.597*** (.200)	−.096 (.211)	.028 (.068)
Sex (Boys)	.208 (.194)	−304 (.208)	.090 (.067)
Ethnicity	.302 (.212)	.152 (.230)	−.027 (.076)
Age	−.001 (.011)	.003 (.012)	−.001 (.004)
Family income	−.158* (.066)	.025 (.071)	−.027 (.023)
Marital conflict	−.002 (.036)	.007 (.040)	−.003 (.013)
Harsh parenting (Boys)	.091* (.035)	.028 (.038)	−.011 (.012)
Harsh parenting change	.018 (.034)	.054 (.036)	−.022 (.012)
Resting RSA (Boys)	1.472 (1.837)	−1.396 (1.939)	.785 (.600)
Sex x Harsh parenting	−.042 (.040)	−.015 (.043)	.005 (.014)
Sex x Resting RSA	−2.052 (2.251)	.383 (2.384)	−.160 (.747)
Harsh parenting x Resting RSA	.071 (.384)	−.449 (.407)	.182 (.127)
Sex x Harsh parenting x Resting RSA	−.256 (.504)	.912* (.435)	−.372* (.169)

Note. Estimates are regression coefficients and based on square root transformations of delinquency scores. SE is the standard error. Sex was coded 0 for boys and 1 for girls. Race was coded 0 for African American and 1 for European American. RSA = respiratory sinus arrhythmia.

^*p < .05.

^**p < .01.

^***p < .001.

The interaction between sex, harsh parenting, and resting RSA predicted both linear and quadratic change over time in delinquency and is depicted in Figures 1A and 1B. This three-way interaction accounted for significant amounts of variance in linear (R² = .10) and quadratic (R² = .11) change over time in delinquency. The highest levels of delinquency were found for children with lower resting RSA who experience high levels of harsh parenting. This effect was most evident for boys, who exhibited significant linear increases in delinquency over time (B = .42, p < .01). Notably, boys who experience high levels of harsh parenting tended to have higher initial levels of delinquency symptoms, but high resting RSA for boys provided partial protection against the development of delinquency. No other linear or quadratic simple slopes were significant for boys.

Figure 1.

Figure 1A. Interaction between harsh parenting and resting RSA predicts development of delinquency for boys.

Figure 1B. Interaction between harsh parenting and resting RSA predicts development of delinquency for girls.

Girls tended to show less variability in change over time in delinquency symptoms. In contrast to the pattern for boys, girls with lower resting RSA who experience high levels of harsh parenting exhibited high initial levels of delinquency and a significant linear decrease in symptoms (B = −.35, p = .04). The only other significant slopes for girls were found for those who have higher resting RSA and who experience lower levels of harsh parenting. These girls exhibited a significant linear decrease in symptoms (B = −.63, p < .01) along with a significant positive quadratic slope (B = .25, p = .01) that reversed this decline. By age 16 girls with higher resting RSA and low levels of harsh parenting evidenced delinquency symptoms on par with girls who experienced high levels of harsh parenting regardless of their levels of resting RSA.

Predictors of Age 16 Substance Use

Age 16 substance use was predicted from the same independent variables used in the model predicting the development of delinquency. Most notably, we were interested in the interactions between harsh parenting, resting RSA, and sex as predictors of substance use. Four logistic regressions were used to predict whether adolescents had engaged in tobacco use, alcohol use, marijuana use, and other drug use that included illicit substances such as cocaine and methamphetamine. Results from these logistic regressions are presented in Table 3. It is important to note that proportions of substance use were relatively low overall, ranging from 9% of adolescents having used other drugs, to 44% of adolescents having used alcohol. These relatively low rates of use are reflected in the predicted probabilities of substance use in our figures of results.

Table 3.

Logistic Regression Results for Prediction of Age 16 Substance Use

	Tobacco use		Alcohol use		Marijuana use		Other drug use
	Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE
Intercept	−1.482***	.222	−.505	.375	−2.206***	.622	−2.500***	.750
Sex (Boys)	1.744*	.982	.476	.444	1.173	.707	.727	.882
Ethnicity	−.712	.641	−.390	.471	−.976	.691	−1.651	.916
Age	.054*	.032	.034	.023	.072*	.035	−.023	.039
Family income	−.219	.197	−.242	.152	−.616**	.222	−.329	.269
Marital conflict	.061	.102	.040	.077	.055	.114	.272*	.122
Harsh parenting (Boys)	−.142	.247	−.029	.112	.123	.174	.106	.244
Harsh parenting change	.075	.100	.071	.074	.102	.108	−.167	.157
Resting RSA (Boys)	8.081	6.431	−3.855	3.806	3.587	5.271	−2.027	8.216
Sex x Harsh parenting	.217	.242	.103	.105	−.100	.162	−.103	.221
Sex x Resting RSA	.391	7.470	7.925	5.235	4.816*	2.12	10.852	10.062
Harsh parenting x Resting RSA	−.304	1.846	−1.384	1.183	−2.19**	.988	−1.872*	.961
Sex x Harsh parenting x Resting RSA	−.001	2.024	1.763	1.441	2.268	1.831	5.117*	2.448

Note. Estimates are logistic regression coefficients predicting membership in group coded as 1; levels coded as 1 indicate youth have participated in the activity (e.g., has used tobacco, alcohol, marijuana, or other drugs). SE is the standard error. Sex was coded 0 for boys and 1 for girls. Race was coded 0 for African American and 1 for European American. RSA = respiratory sinus arrhythmia.

^*p < .05.

^**p < .01.

^***p < .001.

The only predictors of tobacco use were age and sex, with children who were older at the first assessment and girls being more likely to have used tobacco in adolescence. No independent variables predicted adolescent alcohol use.

Age was linked to increased likelihood of marijuana use, and income was linked to decreased likelihood of its use. Additionally, being female and higher resting RSA were related to greater likelihood of marijuana use; this main effect occurred in the context of two interactions. Surprisingly, the interaction between sex and resting RSA indicated that likelihood of marijuana use increased as resting RSA increased, but only for girls (Figure 2A). The interaction between harsh parenting and resting RSA is depicted in Figure 2B and shows that harsh parenting was related to increasing probability of having used marijuana, but only for children with lower resting RSA.

Figure 2.

Figure 2A. Interaction between sex and resting RSA predicts odds of marijuana use at age 16.

Figure 2B. Interaction between harsh parenting and resting RSA predicts odds of marijuana use at age 16.

Higher levels of marital conflict at age 8 increased the probability of other drug use at age 16. A two-way interaction between harsh parenting and resting RSA predicted other drug use, but this was subsumed by a significant three-way interaction between harsh parenting, resting RSA, and sex. This interaction is depicted in Figure 3 and indicates that the combination of high harsh parenting and low resting RSA was related to increased probability of other drug use, especially for boys. This same relationship held for girls, though it was somewhat less strong.

Figure 3.

Interaction between sex, harsh parenting, and resting RSA predicts odds of other drug use at age 16.

Predictors of development of resting RSA

Results from this model can be found in Table 4. Ethnicity was related to resting RSA at age 8, with European American children having lower resting RSA, but was not related to change in resting RSA over time. We also found evidence for an interaction between sex and RSA withdrawal. Plotting this interaction showed that less RSA withdrawal was predictive of increases in resting RSA over time only for girls.

Table 4.

Longitudinal Model Results for Prediction of Resting RSA

	Resting RSA
	Intercept B (SE)	Linear slope B (SE)	Quadratic slope B (SE)
Estimate	.181*** (.012)	−.002 (.015)	.001 (.005)
Sex (Boys)	−.008 (.012)	.002 (.015)	.001 (.005)
Ethnicity	−.031* (.013)	.017 (.016)	−.003 (.005)
Age	.001 (.001)	−.001 (.001)	−.001 (.001)
Family income	−.004 (.004)	.001 (.005)	−.001 (.001)
Marital conflict	−.001 (.002)	−.002 (.003)	.001 (.001)
Harsh parenting (Boys)	−.002 (.002)	.001 (.002)	.001 (.001)
Harsh parenting change	.002 (.002)	−.001 (.003)	.001 (.001)
RSA withdrawal (Boys)	−.137 (.132)	.039 (.158)	.012 (.050)
Sex x Harsh parenting	.004 (.002)	.001 (.003)	−.001 (.001)
Sex x RSA withdrawal	−.705*** (.189)	.601** (.227)	−.133 (.073)
Harsh parenting x RSA withdrawal	−.024 (.024)	.090*** (.028)	−.028*** (.009)
Sex x Harsh parenting x RSA withdrawal	−.067 (.037)	−.013 (.046)	.010 (.015)

Note. Estimates are regression coefficients. SE is the standard error. Sex was coded 0 for boys and 1 for girls. Race was coded 0 for African American and 1 for European American. RSA = respiratory sinus arrhythmia.

^*p < .05;

^**p < .01;

^***p < .001.

Of greatest relevance to our hypotheses, an interaction between harsh parenting and RSA withdrawal predicted the linear slope (R² = .17) and quadratic slope (R² = .14) of resting RSA. This interaction is depicted in Figure 4. In an example of equifinality, children who experience lower levels of harsh parenting but different levels of RSA withdrawal exhibited different levels of resting RSA at age 8 but ended with similar levels of resting RSA at age 16. The differences in resting RSA between children who experience lower or higher levels of harsh parenting are most pronounced at age 16. Children with greater RSA withdrawal who experienced low levels of harsh parenting exhibited high initial resting RSA that did not show significant linear or quadratic change over time. Conversely, children with less RSA withdrawal who experienced low levels of harsh parenting exhibited moderate resting RSA that showed a significant linear increase (B = .048, p < .01) but not quadratic change over time. By contrast, children who experienced higher levels of harsh parenting exhibited the lowest levels of resting RSA; these differences are most clearly evident at ages 10 and 16. Children with greater RSA withdrawal at age 8 who experienced high levels of harsh parenting exhibited the highest resting RSA at age 8 and a moderately significant linear decline (B = −.023, p = .09) without any significant quadratic change. Children with less RSA withdrawal at age 8 who experienced high levels of harsh parenting exhibited the lowest resting RSA at age 8 which remained relatively low over the period of study due to a significant linear slope (B = .060, p < .01) that was tempered by a significant quadratic slope (B = −.012, p = .02) which slowed and then reversed this increase at later time points.

Figure 4.

Interaction between harsh parenting and RSA withdrawal predicts development of resting RSA.

Discussion

We addressed the development of delinquency from middle childhood to adolescence and levels of substance use in adolescence from a biosocial perspective. Recent work has strongly emphasized the importance of interactions between biological and social factors for understanding adaptive and maladaptive developmental trajectories, and researchers have begun to advocate this approach not just in the identification of individuals at risk, but also in the design and implementation of intervention and prevention programs (Beauchaine, Neuhaus, Brenner, & Gatzke-Kopp, 2008; Cicchetti & Gunnar, 2008; van Goozen & Fairchild, 2008). This timely special issue contributes to moving research in this area forward by focusing on stress sensitivity and its role in the etiology, development, and maintenance of psychopathology. We focused specifically on the PNS because of its theoretical and empirical importance in social affiliative behavior and self-regulation (Porges, 2003; 2007; Thayer & Lane, 2000). Our contribution to the special issue highlights two aspects of stress sensitivity in relation to psychopathology. First, we evaluated biosocial interactions by examining the role of resting RSA in the context of harsh parenting to predict changes over time in delinquency from age 8 to age 16, as well as substance use at age 16 (Aim 1). Second, we evaluated the development of resting RSA as predicted by harsh parenting and RSA withdrawal (Aim 2). Thus, our first question addressed resting RSA as a biological risk, or sensitivity, factor for developmental problems in the context of environmental stress and our second question addressed calibration of this potential biomarker of stress sensitivity over time. Consistent with calls for a better understanding of the role of sex in the development of heterotypic psychopathologies (Zahn-Waxler, Shirtcliff, & Marceau, 2008), we also explored sex as a potential moderator in these biosocial models (Aim 3).

Delinquency and Substance Use

In our first set of analyses we found a complex three-way interaction between sex, resting RSA, and harsh parenting in predicting the development of delinquency. This interaction showed that in contexts of harsh parenting and low resting RSA, the developmental form of delinquency was different for the most problematic or at-risk boys (who showed linear increases) and girls (who showed relatively high and stable delinquency symptoms between ages 10 and 16). Additionally, in contexts of harsh parenting, high resting RSA partially protected boys from the development of delinquency problems in middle childhood. This protective effect seemed to fade somewhat during the adolescent transition. The expression of dysregulation in response to family stress becomes increasingly differentiated by sex during adolescence (Zahn-Waxler et al., 2008), and was manifested here as increased delinquency problems and odds of drug use for boys. To a degree, however, the risk factors of low resting RSA and harsh parenting applied to girls’ delinquency and drug use as well. The extension of this study into adolescence thus revealed some novel developmental effects that have not consistently been found in studies of early or middle childhood. These findings echo those of a handful of other studies tracking individuals through the transition from childhood to adolescence that suggest the functionality of stress response systems and their implications for the development of psychopathology change during adolescence (Gunnar, Wewerka, Frenn, Long, & Griggs, 2009).

Our findings are largely consistent with research on other domains of family stress in childhood, most notably research focused on marital conflict. As reviewed by El-Sheikh and Erath (2011), lower resting RSA has been found to serve as a risk factor for both internalizing and externalizing problems in contexts of marital conflict, with some evidence that sex serves as an additional moderator when predicting internalizing versus externalizing problems. For example, in homes marked by high levels of marital conflict, boys (ages 8 to 10) with lower resting RSA showed the greatest increases in delinquency problems over time (El-Sheikh & Hinnant, 2011). Though fewer in number, several studies have found that harsh family contexts interact with lower sympathetic nervous system activity to predict externalizing problems (Bubier, Drabick, & Breiner, 2009; Erath, El-Sheikh, Hinnant, & Cummings, 2011; Shannon, Beauchaine, Brenner, Neuhaus, & Gatzke-Kopp, 2007). In the current study, we extended the study of biosocial interactions to harsh parenting and PNS activity, accounting for effects of marital conflict and changes in harsh parenting.

Risky behaviors like illicit substance use often go hand-in-hand with adolescent delinquency problems (Angold et al., 1999). Research relating autonomic function to substance use is highly novel and may give important insights into the biological underpinnings of substance use initiation and eventual addiction. The only study to our knowledge to address the topic found that blunted sympathetic nervous system reactivity, indexed by pre-ejection period to monetary reward, was related to increased probability of alcohol use for young teens (Brenner & Beauchaine, 2011). In addition to the link between lower resting RSA and delinquency, we found interactions indicating that lower resting RSA was a risk factor for marijuana and other drug use but was unrelated to probability of alcohol and tobacco use. More specifically, in contexts of harsh parenting, low resting RSA predicted increased probability of marijuana use and other drug use (encompassing several of the more common types of “hard” drugs) at age 16. The interaction between harsh parenting and low resting RSA as a predictor of likelihood of other drug use was particularly strong for boys. These findings are consistent with our initial hypotheses.

Two findings that were contrary to expectations were also found. The first was the positive relation between resting RSA and likelihood of marijuana use for girls. The second was the increase in delinquency symptoms and moderately elevated probability of other drug use at age 16 for girls with higher resting RSA who experience low levels of harsh parenting. One possible explanation for these unexpected findings is resting RSA’s positive relation to temperamental measures of novelty-seeking and exploratory behavior (Blandon, Calkins, Keane, & O’Brien, 2010; Dietrich et al., 2009); stronger drive to seek out new experiences may place girls in peer contexts where they have opportunities to engage in risky behavior and experiment with substance use, though we are unsure why this would not be relevant for boys. It is also important to note the low to moderate rates of substance use in the sample; it is possible that these findings would not generalize well to populations containing teens with significantly higher rates of substance use. It is also worth noting that use of different substances is correlated, yet our analyses assume independence. Specificity of effects to substance use outcomes that accounts for this covariance should be examined in future studies.

Findings for our first aim contribute new knowledge of biosocial interactions on risk for psychopathology in several ways. First, we extend upon prior work on the moderating role of family stress beyond marital conflict (El-Sheikh et al., 2011; El-Sheikh & Hinnant, 2011) to test whether harsh parenting has similar interactive effects with PNS activity on externalizing problems. Second, we extend the study of biosocial interactions across the transition from childhood to adolescence to give novel insights into how the influences of risk and protective factors may change during this transition, something few studies have done. Overall, findings illustrate that lower resting RSA is related to both behavior problems and substance use, especially in contexts of harsh parenting. Although there are many potential explanations for these links, at a broad level these processes may involve teens with poor self-regulatory capacities in contexts of harsh and rejecting parenting seeking out peer groups that promote rebellious and risk-taking behaviors (Chung & Steinberg, 2006; Ge, Brody, Conger, Simons, & Murry, 2002; Weaver & Prelow, 2005). Testing this conjecture is not possible with our current data, but it is quite possible that transactions between harsh family environments and developing stress response systems over time exacerbate self-regulatory problems.

Having established that low resting RSA is a risk factor for developmental problems in contexts of high family stress through results presented here and as seen in other empirical work (reviewed in El-Sheikh & Erath, 2011; Obradović, 2012), our second research question addressed the development of resting RSA. In addition to the importance of illuminating the biological risk factors in contexts of stressful environments, researchers also emphasize the importance of understanding the processes that contribute to biomarkers of risk or stress sensitivity (Juster et al., 2011; Lupien, McEwen, Gunnar, & Heim, 2009; Obradović, 2012). Thus, our second contribution addressed the altered functionality of stress response systems by predicting the development of resting RSA as a function of interactions between harsh parenting and RSA withdrawal to stress.

In a prior paper with children ranging in age between 8 and 10 years (some of them participated in study waves 1 to 3 of the current investigation), we presented evidence that high levels of marital conflict or increases in marital conflict over time in combination with greater RSA withdrawal to stress predicted declining resting RSA over time for boys (El-Sheikh & Hinnant, 2011). Controlling for marital conflict and change in harsh parenting, in the current paper we found that higher levels of initial harsh parenting in combination with greater RSA withdrawal predicted significant declines in resting RSA through age 16. Thus, harsh parenting and marital conflict seem to elicit similar calibrations in regards to PNS activity at rest.

These results are consistent with the allostatic load framework (McEwen & Stellar, 1993): Repeated activation of physiological stress responses (e.g., RSA withdrawal) in the context of intense environmental stress (e.g., harsh parenting) may damage or disrupt these systems over time (e.g., incomplete recovery producing low resting RSA and decreased threshold for stress responding), potentially yielding increased sensitivity to stress. A related explanation for these findings incorporates the concept of “repeated hits” from Juster and colleagues (2011) along with physiological recovery from stress. Chronic, repeated stressors such as experiences with verbal or physical harsh parenting may elicit RSA withdrawal which is adaptive, but extended RSA withdrawal may hinder full recovery to prior resting levels, over time resulting in a decline in resting RSA. Lower levels of resting RSA may create something like a “floor effect,” undermining the capacity for social engagement and self-regulation while imposing limits on subsequent RSA withdrawal to future stressors (Hinnant, El-Sheikh, Keiley, & Buckhalt, 2013). Decreased resting RSA may also decrease the threshold for activation of other stress response systems that are normally reserved for coping with more serious threats (i.e., fight-flight responding commonly associated with sympathetic nervous system and hypothalamic pituitary adrenal activation; Beauchaine, 2001). Consideration of other stress response systems concurrently would do much to illuminate this area.

Related to the point above, future research on stress responsivity and sensitization should continue to focus efforts on elucidating the developmental processes leading to hyper- and hypoarousal of physiological systems, addressing both resting states and reactivity to stress. It is encouraging to see that research from multiple and sometimes competing theoretical perspectives is beginning to converge on these issues (for examples see, Davies, Martin, Cicchetti, & Hentges, 2012; Del Giudice, Hinnant, Ellis, & El-Sheikh, 2012). Additionally, researchers from multiple perspectives (Ellis, Del Giudice, & Shirtcliff, 2013; El-Sheikh & Erath, 2011; Hostinar & Gunnar, 2013; Lupien et al., 2009) strongly emphasize the need for longitudinal studies that can assess the development of stress response systems at sensitive periods or transition points in the lifespan. The present findings shed some light on biosocial interactions during one such transition point in the prediction of change over time in delinquency and adolescent substance use while also exploring changes in resting PNS activity. As is common with research, the results contribute to our understanding of developmental issues while raising many more important questions about how familial stress affects physiological systems, and how the two jointly influence developmental trajectories leading to multifinal outcomes.

Results of the present study should be considered in the context of several limitations. Results based on community samples do not necessarily generalize to higher-risk or clinical populations. For example, it is possible that problem behaviors and physiological dysregulation emerge differently in the context of harsh parenting among children with early-starting or adolescent-limited conduct problems. A second limitation is that the study enrolled children from two-parent homes. While we have found evidence for significant variability in family stress in the sample, children from single parent homes often experience higher levels of stress due to their family composition and may constitute a higher risk sample. Findings from such a sample may have yielded different patterns of results. Although we controlled marital conflict, another limitation is our focus on single indices of family stress (harsh parenting) and physiological dysregulation (RSA). Clearly, a wider variety of social and biological variables shape the development of problem behaviors, and future research with the power to examine multiple social and biological predictors will advance the existing literature. Additionally, more complete tests of stress sensitization models should examine how dynamic changes in predictors of stress are related to stress sensitization over time. Despite these limitations, the present study advances understanding of the development of stress sensitivity, as reflected in PNS activity, as well as its implications for problem behaviors in adolescence.