Developmental Connections Between Socioeconomic Status, Self-Regulation, and Adult Externalizing Problems

Abstract

Children from low socioeconomic status (SES) backgrounds are at particularly heightened risk for developing later externalizing problems. A large body of research has suggested an important role for self-regulation in this developmental linkage. Self-regulation has been conceptualized as a mediator as well as a moderator of these connections. Using data from the Child Development Project (CDP, N = 585), we probe these contrasting (mediating/moderating) conceptualizations, using both Frequentist and Bayesian statistical approaches, in the linkage between early SES and later externalizing problems in a multi-decade longitudinal study. Connecting early socioeconomic status, physiology (i.e., heart rate reactivity) and inhibitory control (a Stroop task) in adolescence, and externalizing symptomatology in early adulthood, we found the relation between SES and externalizing problems was moderated by multiple facets of self-regulation. Participants from lower early SES backgrounds, who also had high heart rate reactivity and lower inhibitory control, had elevated levels of externalizing problems in adulthood relative to those with low heart rate reactivity and better inhibitory control. Such patterns persisted after controlling for externalizing problems earlier in life. The present results may aid in understanding the combinations of factors that contribute to the development of externalizing psychopathology in economically marginalized youth.

Introduction

Self-regulation can be defined as the capacity to achieve goal-directed behavior through moderating emotional, cognitive, and physical actions (Bell & Deater-Deckard, 2007; Nigg, 2017). This capacity has been operationalized through constructs including executive functioning, cognitive control, and risk-taking. Regulatory problems often predict later challenges with physical health, substance dependence, personal finances, and criminal offending outcomes (Moffitt et al., 2011; Palacios-Barrios & Hanson, 2019; Rothbart et al., 2000). Influential models of self-regulation argue that self-regulation has “top-down” and “bottom-up” components (Nigg, 2017). Top-down processes are characterized by the utilization of executive functions, including deployment of attentional and inhibitory control (Baumeister et al., 2007; Hofmann et al., 2012). In contrast, bottom-up processes center on physiological reactivity toward salient stimuli, with measures of heart rate being commonly used (Holzman & Bridgett, 2017; Segerstrom & Nes, 2007). Bottom-up processes may be both targets of and sources of regulation, interfering with, but also supporting top-down elements. The present study considered the developmental implications of two markers of self-regulation in the context of the possible impact of children’s families’ economic background.

Socioeconomic Status, Mental Health, and Self-Regulation

While likely to influence multiple developmental competencies, self-regulation has been a major focus for research groups examining the mental health impacts of poverty and lower socioeconomic status (SES). Multiple meta-analyses underscore that lower childhood SES is related to later psychopathology (Peverill et al., 2021; Piotrowska et al., 2015). This is true for multiple different metrics of SES, including household income, parental education, and subjective perceptions of social status. SES is reliably connected to increased risk for subsequent aggression, oppositional behavior, and other forms of externalizing problems. It is likely that the multiple risk factors associated with lower socioeconomic status and economic marginalization underlie these connections; and in attempting to unpack these links, many research groups have suggested SES-related risk factors (i.e., stress exposure) may impinge upon self-regulatory processes and eventually cascade to poor mental health (Blair & Raver, 2012, 2015; Evans & Kim, 2013).

Connected to these ideas, psychobiological work suggests low childhood SES and exposure to poverty are related to alterations in different components of self-regulation, including executive functioning and physiological reactivity (i.e., heart rate variability, cardiovascular efficiency; Boylan et al., 2016). For example, a meta-analysis of 8,760 children found that lower SES and exposure to poverty were related to poorer executive functions (EFs, one facet of “top-down” regulation, r= .28; Lawson et al., 2018). This was seen for overall/latent constructs of EFs, as well as between SES and more specific EF tasks (e.g., working memory, inhibitory control). Similarly, lower SES and poverty have been associated with multiple indices of physiological reactivity, including salivary cortisol dysregulation in childhood (Blair et al., 2013; Lupien et al., 2001) and greater diastolic blood pressure and heart rate reactivity in adolescence (Chen et al., 2004; McGrath et al., 2006).

However, there are complications in the results. Research has also noted that externalizing behaviors are related to alterations in EFs and heart rate reactivity in adolescence, but these metrics of self-regulation do not always differ by SES groups (Perry et al., 2018). Furthermore, it is important to note that most children developing in lower SES contexts will not develop externalizing problems. More needs to be learned about when lower SES conditions do and do not become connected to externalizing outcomes (Blair & Raver, 2015; Evans & Kim, 2012).

Improving Understanding of Associations Between Self-Regulatory Processes, Socioeconomic Status, and Externalizing Behaviors

Gaps in understanding links between SES, self-regulation, and later developmental outcomes are in part due to: 1) Disagreement about whether self-regulation may be a potential mediator or moderator of SES-externalizing links; 2) Limited focus on the dynamic interplay between different facets of self-regulation (e.g., “top-down”; “bottom-up”); and 3) Complex transactions between developmental contexts (i.e., lower SES and exposure to poverty) and facets of self-regulation.

Related to mediating or moderating pathways, disparate perspectives have emerged for the role of self-regulation in relation to the developmental challenges seen after exposure to lower SES conditions (Blair & Raver, 2012; Doan et al., 2012; Raver et al., 2011). Lower SES is related to lower regulatory control; and independent of SES, poorer regulatory skills are correlated with multiple mental health challenges (Snyder et al., 2019; Zelazo, 2020). Two studies directly testing if variations in self-regulation mediate connections between SES and externalizing problems have found support for this idea (Fatima & Sheikh, 2014; Granero et al., 2015); however, other research studies have noted that low-SES youth with strong regulatory skills have better psychosocial outcomes (e.g., lower rates of psychopathology; better academic functioning) than those with lower executive functions or higher physiological reactivity. In this work, facets of self-regulation (i.e., EFs; salivary cortisol) were not directly impacted by exposure to poverty and lower-SES conditions, but instead buffered against the developmental challenges commonly seen in these contexts (Brown et al., 2019; Fernald & Gunnar, 2009; Flouri et al., 2014; Masten et al., 2012; Staufenbiel et al., 2013). As such, self-regulatory abilities are often framed as a potential resiliency factor, providing protective effects against the stressors associated with lower SES and poverty (Zelazo, 2020). Interestingly, few projects have attempted to examine both potential moderating and mediating pathways within the same sample to aid in adjudicating this debate.

Regarding different facets of self-regulation, limited work has incorporated assessments of bottom-up and top-down self-regulation, with a specific eye toward interactions between these elements. Multiple prominent theoretical models (e.g., Eisenberg & Fabes, 1992; Rothbart et al., 2000) posit that there are dynamic interactions between top-down (regulatory processes, such as inhibitory control) and bottom-up (reactivity) self-regulatory processes. Furthermore, multiple empirical reports have found externalizing types of behaviors were particularly high for individuals both low in regulation and high in reactivity. For example, Eisenberg and colleagues (1994) found that children high in negative emotional intensity and low in top-down regulation were less constructive compared to other children when dealing with anger inducing situations. Over longer time scales and using multiple informants of child behavior, similar interactive patterns have been noted with higher problem behavior being predicted by the interaction of emotional reactivity and regulation (i.e., high reactivity X low regulation) (Eisenberg et al., 1996, 2000; Morris et al., 2014). Additional support for the importance of this interaction comes from school-based interventions that found early childhood curriculum interventions aimed at bolstering self-regulation improved inhibitory control, but also led to changes in physiological reactivity (i.e., salivary cortisol; alpha amylase) (Blair & Raver, 2014; Raver et al., 2011). As such, interactions between these different components of self-regulation (e.g., inhibitory control; heart rate reactivity) may be important to consider and may aid in explaining different developmental competencies.

Finally, evidence suggests there may be complex interactions between developmental contexts (i.e., poverty and lower SES) and facets of self-regulation (i.e., reactivity and regulation). Many of the inter-related elements of self-regulation are strongly influenced by developmental context and may more cogently fit within ecological, multi-systems approaches to youth development (Calkins, 2010; Ellis et al., 2020; Masten et al., 2021; Thompson, 2011). Put another way, self-regulation may not be uniformly beneficial for all youth. Transactions between factors and context produce adaptive (or maladaptive) outcomes. Multisystem theories consider varying levels at which development occurs (i.e., at the individual; family), and the interplays between multiple factors at different “levels of the individual” all carry weight as factors of risk and resilience in relation to the development of various problem behaviors (Sameroff, 2010). In thinking about SES and externalizing problems, multisystem theories would posit that having moderate or high sensitivity to stressors does not uniformly put an individual at an advantage or a disadvantage. Instead, when individuals encounter stressful environments, such as those common in lower SES conditions, this may direct their development toward adaptive strategies and behavioral change (Ellis et al., 2020). This change occurs even if strategies are not beneficial in all life contexts or in the long-term (Ellis et al., 2011). Support for these ideas in relation to SES has come from multiple projects finding that relations between self-regulation and SES are often non-linear (Obradović, 2016). For example, investigations examining delay of gratification, another operationalization of self-regulation, have noted that children from lower SES backgrounds often have more challenges in waiting for delayed rewards (Evans & English, 2002; Raver et al., 2011). Notably, this has been linked to elevated heart rate reactivity and vagal tone (Sturge-Apple et al., 2016). However, data suggest that higher vagal tone is associated with longer delay of gratification in middle-class households, but the opposite pattern in low-SES contexts (Obradović et al., 2016) Such non-linear relations support ideas from multisystem perspectives of development and underscore SES’ role as an important contextual factor for developmental cascades. As such, it may be important to think about interactions between not only top-down and bottom-up facets up self-regulation, but also developmental contexts, and the confluence of these factors in predicting developmental outcomes.

Self-Regulation During Adolescence

We chose to address questions about self-regulation and externalizing behavior problems in adolescence. This era is marked by changing physiology, changes in behavioral regulation, and increases in mental health issues. Self-regulatory processes begin to differentiate in late childhood and early adolescence, allowing individuals to deal with both normative and stressful experiences (Crone & Dahl, 2012). Cognitive regulation (such as inhibitory control) and reactivity processes play a critical role in adolescents’ abilities to navigate changing emotional and social contexts, particularly in stressful circumstances. Commonly studied facets of regulation that mature during the adolescent transition are goal directed behaviors, planning, and inhibitory control, which allow individuals to evaluate and alter responses to environmental circumstances (Gestsdottir & Lerner, 2008). Strong self-regulation strategies during adolescence have been linked with improved academic, social, and emotional outcomes when faced with stressful situations, and lower sexual risk-taking, delinquency, and substance use (Pandey et al., 2018). Additional research is needed to understand developmental antecedents and consequences of self-regulation.

The Current Study

To overcome many of the conceptual and methodological inconsistencies reviewed above, we examined associations among measures of early childhood developmental context (i.e., exposure to lower SES), facets of adolescent self-regulation, and adult psychopathology, using data from a unique longitudinal project that spans from early childhood into adulthood (the Child Development Project, CDP). We focused on regulatory processes during adolescence, as research during this developmental epoch is limited.

Using CDP’s multi-method design, we aimed to investigate connections between exposure to early low SES, facets of top-down and bottom-up self-regulation, and measures of adult externalizing psychopathology. The current study tests two competing hypotheses (e.g., mediation; moderation) using Frequentist and Bayesian statistical approaches. First, based on the idea that lower SES and poverty are associated with greater externalizing problems, we tested if two different measures of self-regulation, a task based inhibitory control and physiological reactivity, play a mediating role in SES-externalizing links. We predicted that lower childhood SES will be associated with lower inhibitory control and greater physiological reactivity, and that these differences would statistically explain links between SES and externalizing behaviors. Second, and in contrast to mediation models, we examined if the interaction of self-regulation and developmental context would be related to externalizing problems. Motivated by multisystem models of development, we specifically posited that lower top-down control and greater physiological reactivity would relate to externalizing symptoms, but only for those exposed to lower SES conditions early in development.

Methods

Data were drawn from the CDP, which recruited children enrolling in kindergarten in 1987 and 1988 at three-sites (Knoxville and Nashville, TN, and Bloomington, IN). The total sample included 585 kindergartners (52% Male, 48% Female). The sample varied in socioeconomic status using the Hollingshead index (M= 39.61, SD = 13.97, Range: 11–66), and included 81.5% White, 16.6% Black, and 1.9% other. Participants were followed prospectively for the proceeding 20 years, and data were collected approximately once a year. For these analyses, we utilized data from when participants were in kindergarten (demographics and socioeconomic status), 10^th and 11^th grade (EF and heart rate reactivity), and age 26 (externalizing behaviors). This allowed us to connect early environmental circumstances (in early childhood) and aspects of self-regulation during a developmental transition (late adolescence), with more stable adult outcomes. At age 26, 83.90% of participants provided data on psychopathology, and these individuals were statistically similar to the full sample; descriptive statistics for relevant variables are noted in Table 1. Related to missingness, parametric (Hawkins, 1981) and non-parametric testing (Jamshidian & Jalal, 2010) suggest data are missing completely at random (with homoscedasticity among groups of cases with missing data patterns, all ps>0.399).

Table 1.

Descriptive Statistics for All Participants

	N	M	SD	Range
Early Socioeconomic Status (Hollingshead Index at Age 6)	559	39.59	13.96	11–66
Later Socioeconomic Status (Hollingshead Index at Age 17)	445	39.26	13.86	6–66
Early Externalizing Problems (CBCL Externalizing at Age 6)	567	11.51	7.02	0–39
Later Externalizing Problems (ASR Externalizing at Age 26)	469	7.74	6.20	0–37
Inhibitory Control (Stroop Errors During Adolescence)	404	2.20	2.20	0–12
Heart Rate Reactivity (in Adolescence)	386	74.16	11.07	47.85–102.48

Developmental Context: Socioeconomic Status

At the start of the study, when participants were in kindergarten, their parents/caregivers reported on household education and occupation. These measures were then combined to produce a composite of socioeconomic status (Hollingshead, 1975). Hollingshead Index raw scores range from 11 to 66, with higher scores representing higher parental education and occupational prestige (critical elements of SES). At this earliest study time point, average Hollingshead was 39.61 corresponding to skilled craftsmen, clerical, and sales workers. In addition, when participants were in high school, SES levels were collected again. Early SES scores were the primary variable of interest for this study’s analyses; however, later SES levels were used as covariates. This covariate inclusion was to examine the specificity of SES effects, as growing research suggests the importance of this early childhood context (Heckman, 2006).

Top-Down Self-Regulation: Inhibitory Control

When participants were 16 years old, executive functioning, specifically inhibitory control, was measured in participants via a Stroop task. This measured the participants’ ability to exercise control over prepotent behavioral responses during the presentation of pictures and words. Participants were shown sets of four cards. The first card contained only the word for an animal, such as “dog,” and the participant was asked to say the word on the card. For the second card, participants saw a picture of an animal, but no words (e.g., a drawing of a dog). Then, the participant was asked to say the name of the animal in the picture. The third card contained both a picture of an animal and the name of a different animal (e.g., a picture of a cat with the word “Pig”), and the participant was asked to say the name of the drawn animal but not the word on the card. For the last card of each set, participants saw a picture of an animal and a nonsense word (e.g., a picture of a bear and the nonsense word “Fyt”). Participants were asked to ignore the nonsense word on the card and say the name of the animal in the picture. The number of errors on this task was used to measure top-down control. If participants did not say the name of the animal in the picture (e.g., saying the word on the card when it did not match the picture, failing to answer, or saying the name of another animal), this was counted as an error.

Bottom-Up Reactivity: Heart Rate Reactivity

When participants were approximately 16 years old, psychophysiological data were collected, specifically heart rate levels of participants. Participants were asked to apply self-adhesive heart rate sensors to their breastbones after being instructed on the process by a research assistant. As described in greater detail elsewhere (Crozier et al., 2008), heart rate data were collected at baseline and while participants were exposed to different video vignettes. Heart rate data acted as a proxy of autonomic nervous system (ANS) response as the vignettes were played. These vignettes were designed to evoke ambiguous, nonaggressive, or hostile responses to stimuli. The stories depicted age-appropriate events, such as asking someone to a dance or needing to sharpen a pencil during class, and the stories featured adolescents of similar age to the participants. Participants watched 18 video segments across six videos (three segments per video). After each story, participants were asked to answer various questions about the vignettes, including summarizing what occurred and how they would act in the depicted situation. They also indicated whether the other adolescent in the video was responding aggressively (e.g., “Being mean,” “Not being mean,” “It’s hard to tell,” and “Don’t know”) and whether each provided response was appropriate (e.g., “BAD,” “bad,” “good,” “GOOD”). Physiological data were then processed to obtain metrics of cardiac reactivity. First, physiologically unlikely values of heart rate (<30 bpm, and >150 bpm at rest) were coded as artifact and removed; the average amount of intrasubject artifact was 1.8%. Next, second-by-second heart rate was averaged for rest, as well as during the vignettes depicting different social situations. We then calculated difference scores between rest and different vignettes, subtracting rest from reactive heart rates averages). Based on past work (Crozier et al., 2008), we focused our analyses on the difference score for hostile vignettes versus rest.

Externalizing Symptomatology

When participants were approximately 26 years old, they completed the Adult Self Report (ASR; Achenbach & Rescorla, 2003). This measure was designed to produce information on adaptive functioning and mental health issues, such as rule-breaking, anxious/depressed, and aggressive behavior in adult participants from 18 to 59 years of age. The ASR consists of 123 questions for which participants indicate their level of agreement (e.g., “0-Not True,” “1-Somewhat or Sometimes True,” “2-Very True or Often True”). Twenty-eight items in the externalizing subscale were summed to create a composite (α=.84). Supplemental, sensitivity analyses were also completed controlling for early externalizing behaviors reported by parents, when participants were in kindergarten via the Child Behavior Checklist (Achenbach, 1994).

Analytic Plan

Based on the literature, we were interested in testing two main hypotheses. First, we aimed to see if top-down self-regulatory processes (i.e., inhibitory control) and/or bottom-up reactivity (i.e., heart rate reactivity) during adolescence mediated associations between early SES, and externalizing symptoms in adulthood. This first involved construction of models examining whether early family SES (X) was associated with externalizing symptomatology in adulthood (Y) and if any observed associations were explained by two mediating processes: inhibitory control (M1) and heart rate (M2). Statistical testing of mediation was done by nonparametric bootstrapping, with 95% confidence intervals (CIs) for indirect mediation effects. Second, connected to multisystem theories of development, we examined associations between early SES and self-regulatory processes in relation to externalizing symptoms. For these analyses, we examined the two- and three-way interactions between two self-regulatory processes (inhibitory control; heart rate) and SES. For all analyses, we constructed linear regression models with externalizing behaviors at age 26 entered as the dependent variable, with heart rate and number of inhibitory control errors (and their interactions) entered as the independent variables, and also heart rate, number of inhibitory control errors, and SES (and their interactions) entered as the independent variables. Relevant covariates were included in all models (i.e., sex, site, race). For all significant interaction terms, we then conducted follow-up analyses of the simple slopes of subgroups (the mean of all individuals, + 1 SD of the mean, and −1 SD of the mean) using the “interactions” R library (Long, 2019). Of note, all statistical models used a “list-wise deletion” approach for any missing data and sensitivity analyses used other approaches to missing data approach (e.g., Full Information Maximum Likelihood).

Supplemental (Sensitivity) Analyses

After initial model construction, we completed supplemental analyses using alternative statistical specifications, as well as different strategies for dealing with missing data. Notably, we planned to use Bayesian approaches to test mediation (M1: inhibitory control; M2: heart-rate reactivity) and moderation. Bayesian statistics improved over standard (Frequentist) modeling by having better model accuracy in noisy data. Specifically, Bayesian models do not rely on the common “p value framework” of Frequentist statistics; this may be advantageous as p-values especially can allow researchers to draw strong(er) conclusions based on whether an effect is significant versus not (i.e., p<.05). Furthermore, Bayesian models are often less prone to type I errors and have the ability to speak to both the presence and the absence of effects (Makowski, Ben-Shachar, Chen, et al., 2019). To test mediation, the highest density interval (HDI) and 89% confidence intervals were calculated for Bayesian mediation and moderation effects in keeping with current best practices (Kruschke, 2014; Makowski, Ben-Shachar, & Lüdecke, 2019). The HDI indicates the points of a distribution that are most “credible” and that cover most of the distribution.

Related to covariates, we constructed additional sensitivity models including: a) measures of early externalizing problems; and b) socioeconomic status during adolescence in relation to the mediation and moderation analyses noted above. Finally, as shown in our supplemental materials, we completed additional analyses of interest. These included alternative approaches to dealing with missing data, as well as single mediation models. Related to missing data, all models in the main manuscript (Frequentist; Bayesian) used a list-wise deletion approach for any missing data. We, however, also completed analyses using Full Information Maximum Likelihood methods to deal with missing data, primarily for the measures in adolescence (inhibitory control and heart rate) (Collins et al., 2001). In this supplement, we also included mediation models only examining differences in inhibitory control (M) in explaining links between early SES and later externalizing symptoms.

Results

Association Between Early SES and Later (Adult) Externalizing

In line with many past reports, early family SES, measured around kindergarten, was related to later, adult externalizing problems (β=−0.139, p=0.006). Greater problems were seen in children from lower SES households. A scatterplot of this association is shown in Figure 1. This relation lessened but was still significant when adjusting for early externalizing problems, also measured in kindergarten (β=−0.115, p=0.022). Similar associations were not seen with adult externalizing and later socioeconomic status, measured near the end of high school (β=−0.086, p=0.099). These two effects were significantly different from one another (t=14.75, p<.005), suggesting that early SES is a more robust predictor of later externalizing problems (compared to SES measured later in development). Table 2 shows the bivariate correlations between the primary variables examined in our study.

Figure 1.

Scatterplot showing the association between early SES and adult externalizing problems (p=.006)

Table 2.

Means, standard deviations, and correlations with confidence intervals

Variable	M	SD	1	2	3	4	5	6	7
1. Childhood.SES	39.59	13.96
2. Adolescent.SES	39.26	13.86	.70**
			[.65, .74]
3. Childhood Externalizing	11.51	7.02	−.14**	−.18**
			[−.22, −.06]	[−.27, −.08]
4. Adult.Externalizing	7.74	6.20	−.11*	−.07	.21**
			[−.20, −.02]	[−.17, .03]	[.12, .30]
5. Inhibitory.Control	2.20	2.20	−.18**	−.11*	.09	.17**
			[−.27, −.08]	[−.21, −.00]	[−.01, .19]	[.06, .27]
6. Heart.Rate	74.16	11.07	−.01	.01	−.03	.06	.02
			[−.11, .09]	[−.09, .12]	[−.13, .07]	[−.05, .16]	[−.08, .12]
7. Race	1.21	0.40	−.39**	−.29**	−.00	.03	.30**	−.09
			[−.46, −.32]	[−.38, −.21]	[−.08, .08]	[−.06, .12]	[.21, .39]	[−.19, .01]
8. Sex	1.48	0.50	−.05	−.07	−.04	−.17**	−.02	.12*	.03
			[−.13, .03]	[−.16, .03]	[−.12, .05]	[−.26, −.08]	[−.12, .07]	[.02, .21]	[−.05, .11]

Note. M and SD are used to represent mean and standard deviation, respectively. Values in square brackets indicate the 95% confidence interval for each correlation. The confidence interval is a plausible range of population correlations that could have caused the sample correlation (Cumming, 2014).

^*indicates p < .05.

^**indicates p < .01.

Probing the Mediating Role of Self-Regulation (Hypothesis 1)

Motivated by past work suggesting that differences in self-regulation may mediate connections between SES and later negative developmental outcomes, we examined whether two different facets of self-regulation, top-down inhibitory control or bottom-up heart rate reactivity, statistically explained the links between early SES and later externalizing (as conceptually depicted in Figure 2A). Using a dual mediation framework, the indirect effects connecting SES to inhibitory control, and inhibitory control to externalizing were non-significant (95% CIs =−0.22–0.03, p=0.416), as were indirect effects connecting SES to heart rate reactivity and heart rate reactivity to externalizing (95% CIs=−0.08–0.04, p=0.682). As such, and similar to models focused on inhibitory control only, the combined indirect effects were also non-significant (95% CIs=−0.25–0.04, p=0.384; also see Table 3). Bayesian approaches yielded similar, non-significant, effects (indirect effects for inhibitory control HDI 89% CIs= −0.13–0.02; indirect effects for heart rate reactivity HDI 89%CIs=- −0.08–0.04); this again suggests these variables did not play a mediating role between early SES and later externalizing problems.

Figure 2.

Conceptual models being tested in the present work are shown here. Panel A shows a dual-mediation approach, connecting early SES (X), adult externalizing (Y), and inhibitory control (M1) and heart-rate variability (M2). Panel B depicts a moderation between early SES, inhibitory control, and heart rate variability to predict adult externalizing.

Table 3.

Path Coefficients from Dual Mediation Model

		Standardized
Predictor	DV	Path Values	SE	z	Sig	p	Lower.CI	Upper.CI
Inhibitory Control (a1)	Adult Externalizing	0.125	0.075	1.678		0.093	−0.021	0.271
Heart Rate Reactivity (a2)	Adult Externalizing	0.057	0.050	1.141		0.254	−0.041	0.154
Childhood SES (c)	Adult Externalizing	−0.126	0.060	−2.101	*	0.036	−0.244	−0.008
Childhood SES (b1)	Inhibitory Control	−0.052	0.059	−0.872		0.383	−0.168	0.065
Childhood SES (b2)	Heart Rate Reactivity	−0.020	0.059	−0.344		0.731	−0.136	0.095
a1 *b1	Indirect1	−0.006	0.009	−0.759		0.448	−0.023	0.010
a2 *b2	Indirect2	−0.001	0.003	−0.338		0.735	−0.008	0.006
c+(a1 *b1)+(a2 *b2)	Total	−0.134	0.061	−2.185	*	0.029	−0.254	−0.014

Examining Moderating Influences of Self-Regulation (Hypothesis 2)

Connected to multisystem models (and as conceptually shown in Figure 2B), we also investigated if connections between early SES and adult externalizing problems were moderated by top-down inhibitory control and bottom-up heart rate reactivity. We looked at the three-way interaction among SES, inhibitory control, and heart rate reactivity. There was a significant three-way interaction (β=−0.134, p=0.02). Of note, no two-way interactions (e.g., SES × Inhibitory Control, Inhibitory Control × Heart Rate) were significant (all ps > 0.49). Following up this significant three-way interaction, we completed simple slope analysis for inhibitory control and heart rate variability. Interestingly, at average or above average levels of early SES (mean; +1 SD), all simple slopes were non-significant (all ps > .14). In contrast, for below average levels of early SES (−1 SD), simple slopes for the association of inhibitory control and later externalizing behavior were significant when heart rate variability was higher than average (+1 SD; t=2.7, p=.007). Put another way, when early SES was low and heart rate was high, more errors on the inhibitory control task (indexing poorer self-regulation) were related to more adult externalizing problems. Figure 3 shows a plot of the interactions (Panel A), as well as a Johnson-Neyman plot (Panel B). These models involved ordinary least squares regressions, so missing data were listwise deleted. Similar results were found: 1) using Bayesian approaches (HDI 89% CIs=−1.02– −0.12); and 2) when maximum likelihood estimation approaches were used to deal with missing data (β=−0.132, p=0.022).

Figure 3.

Graphics depicting the significant three-way interaction of early SES, inhibitory control, and heart rate variability in predicting externalizing. Panel A shows inhibitory control on the horizontal axis and externalizing on the vertical axis. For this figure, heart rate variability SES is broken down by +1 SD, average, and −1 SD levels; SES (by −1 SD, average, and +1 SD levels) is shown in the figures subpanels going left to right. Panel B shows a Johnson-Neyman plot noting when the slope of the interaction is significant with teal depicting p<.05 values and pink corresponding to non-significant slopes. SES (by −1 SD, average, and +1 SD levels) is again broken down by subpanels in the figure going left to right.

Speaking to the specificity of these findings, the three-way interaction among early SES, inhibitory control, and heart rate remained unchanged after adjusting our statistical models for early externalizing problems (three-way interaction β=−0.126, p=0.029). Similarly, when controlling for later SES (measured near the end of high school) and early externalizing, the early SES, inhibitory control, and heart rate interaction remained significant (β=−0.131, p=0.029). Interestingly, and speaking to the developmental timing of potential effects, a three-way interaction among high-school age SES, inhibitory control, and heart rate was non-significant (β=−0.052, p=0.41). We also completed exploratory analyses comparing single and dual mediation to moderation using different Bayesian statistics (e.g., Bayes factor, expected log predictive density, leave-one-out Information Criterion) to compare models; these are described in our supplemental materials.

Discussion

Although associations between lower socioeconomic status and later externalizing behaviors have been consistently noted (Comeau & Boyle, 2018; Piotrowska et al., 2015; Slopen et al., 2010), additional research is needed to understand the developmental factors that may explain, or amplify, these connections. In line with this goal, we examined the potential mediating, as well as moderating, role of self-regulation in the association of exposure to low SES environments and externalizing problems. We found that two facets of self-regulation in adolescence interacted with early low SES exposure to predict adult externalizing problems. Specifically, the combination of higher reactive heart rate and poorer inhibitory control in adolescence related to more adult externalizing problems for individuals who were exposed to a low SES environment in early childhood. Put another way—individuals who begin life in lower SES contexts and then have higher bottom-up reactivity and also lower top-down self-regulatory abilities in adolescence, may be at particular risk for increases in externalizing problems as development progresses.

Our findings are notable for multiple reasons. First, and contrary to some theoretical models and past reports, we did not find support for the mediating role of self-regulation in the links between low SES and externalizing symptoms. This was examined with both Frequentist and Bayesian statistical approaches to more richly probe these links. Various self-regulatory abilities have been connected to forms of psychopathology, as well as exposure to poverty and lower SES (Zelazo, 2020); however, there is a surprisingly small number of published papers testing links among these constructs, especially in relation to “top-down” self-regulatory processes, such as executive functions. We examined both “top-down” inhibitory control and “bottom-up” heart-rate reactivity and did not find support for these variables as mediators. This was done using dual-mediation approaches, as well as single-mediation (for only inhibitory control, see Supplemental Materials). Many research groups measure self-regulatory processes and also collect assessments of problem behavior and psychopathology. It would be surprising if these links had not been explored by past research reports. Perhaps null effects have been found, but remain unpublished.

In addition, we found that variations in self-regulatory processes were only related to negative outcomes (externalizing behaviors) in lower SES participants. This may occur for multiple different reasons. Connected to multisystem theories of development, lower SES in childhood may relate to greater stress exposure, and then early stress may cascade to later externalizing issues. With greater stress and potential externalizing vulnerabilities, different patterns of inhibitory control and heart-rate reactivity may be protective against this risk. These ideas relate broadly to diathesis-stress and other latent vulnerability models. Alternatively, and in line with multi-system and multi-level theories of development, it may be that linkages among SES, self-regulation, and mental health are more complex in nature. Put another way, lower SES in childhood and greater stress may direct development towards certain short-term adaptive processes that could also be related to larger-term challenges. Past work (e.g., Obradović et al., 2016) has found stress levels, quantified by cortisol output, moderated the association between family income and cognitive regulation. Those with high cortisol levels tended to perform worse on inhibitory control tasks when family income was low, relative to their low cortisol, low-income peers. In contrast, participants who had high cortisol and a higher family income performed better on regulatory tasks than their high income, low cortisol peers. This connects to recent work finding that better self-regulation relates to better mental health, but actually to poorer physical health and indices of biological aging (Brody et al., 2013; Chen et al., 2015; Miller et al., 2015). As such, researchers may need to think more about the multidimensional nature of self-regulation, especially in relation to different developmental and cultural contexts.

Considering the implications of our findings for applied contexts, self-regulation may be a key target of intervention when children are at risk of later externalizing problems, but only perhaps for certain subgroups of individuals Children from different socioeconomic backgrounds may not respond the same to prevention and intervention programming aimed at boosting self-regulation and other related processes. This will be particularly important to consider as different behavioral interventions targeting self-regulation (e.g., Chicago School Readiness Project) gain traction in school systems (Pace et al., 2019; Ursache et al., 2012).

The present work also underscores the potential importance of understanding the impact of early developmental contexts, such as poverty and lower SES contexts. Indeed, exposure to early poverty may show impacts for decades. However, it is important to realize that the “lived experiences” of poverty and lower SES encompass more than merely low economic resources. It may also be characterized by increased exposure to neighborhood violence, increased family conflict and chaos within the home, and a systemically different set of stimulatory inputs (Evans, 2004). Interestingly, similar interactions between later socioeconomic status, measured in adolescence, and self-regulation were not seen. One should approach interpreting null interaction effects with caution, but the importance of early developmental experience fits well with past econometric work on the outsized impact of interventions in childhood compared to later in development (Heckman, 2006). Failing to reduce early socioeconomic disparities may have particularly prolonged effects, including possibly those beyond young adulthood. Accordingly, steps should be taken on a local and societal level to mitigate poverty’s lasting effects on individuals and reduce the occurrence of costly externalizing problem behaviors.

While this work had multiple strengths (e.g., multi-year longitudinal design, multiple informants, physiological measures), there are a number of important limitations to note. First, there was a fair amount of participant attrition across time-- 386 of our initial 567 participants (~68%) had available heart rate reactivity data. However, while attrition is generally considered a limitation, this may actually be leading to an underestimation of effects as individuals from deeper poverty and/or with greater behavioral problems may be more likely to be missing from the sample. Use of analytic approaches to deal with missing data (e.g., Full Information Maximum Likelihood) found similar interactive effects to those found with listwise deletion and Ordinary Least Squares regression. Second, “bottom up” metrics of self-regulation, like the one we examined here, is a combination of both heart-rate reactivity and regulation that captures the interplay between the sympathetic and parasympathetic nervous systems (Rajendra Acharya et al., 2006). This is not ideal and more complex models that include additional physiological parameters (e.g., respiration; arterial blood pressure) could aid in understanding relations between facets of self-regulation and psychopathology (as discussed in Saul & Valenza, 2021). Additionally, the study was not appropriately powered to look at race- or ethnicity-specific effects. The sample was primarily White (81.5%), and had limited representation of Black, Latinx populations and other communities of color. Similarly, we also were unable to examine potential sex differences. It will be important to study if the interactions discovered here are comparable across males versus females, as externalizing issues are more common in males and recent work suggests developmental interactions between self-regulation and externalizing may vary by sex (Hagan et al., 2020). Even with our moderately large sample size, we would not be appropriately powered to test potential 4-way interactions (SES × Sex × Inhibitory Control × Heart Rate). Future work should focus on recruitment of diverse populations to examine potential interactive effects in more representative and larger cohorts. Third, exposure to lower SES contexts is the primary developmental experience examined in the present work, but obviously, this is a heterogeneous developmental context with varying levels of stress exposure. Lower SES and impoverished environments have more stressors, on average, than more affluent households; however, our conceptualization ignored potential distinct dimensions of this adverse context (e.g., McLaughlin et al., 2014; Young et al., 2020). It would be important in future work to understand if experiences of deprivation, threat, unpredictability, or other experiential dimensions common to lower SES, are uniquely driving the patterns reported here. More broadly, future work would greatly benefit from a more comprehensive measure of stress exposure to better capture the intricacies of the “lived experiences” of poverty.

Limitations notwithstanding, we find support for self-regulation as a moderator in explaining connections between SES and externalizing psychopathology. In line with multi-system models of development, we found that early levels of SES demonstrated differential effects on later externalizing behavior based on levels of self-regulation. We believe this framework can help deepen our understanding of where and how to intervene to reduce externalizing behaviors. Fundamentally, we hope that this work can be used to help promote psychological health across development, regardless of starting socioeconomic status and levels of self-regulatory abilities.

Research Highlights:

• Childhood socioeconomic status was related to externalizing symptoms 20 years later

• Adolescent self-regulation did not mediate links between SES and externalizing

• Youth from low SES households with low inhibitory control and high physiological reactivity showed greater externalizing issues in adulthood