Reinforcement sensitivity theory and externalizing problems across early adolescence: Testing within-person reciprocal associations

Abstract

Temperament and externalizing problems are closely linked, but research on how they co-develop across adolescence remains sparse and equivocal. Reinforcement sensitivity theory (RST) provides a useful framework for understanding temperament and externalizing problems associations. During adolescence, oppositional problems are posited to be linked to an overactive behavioral approach system (BAS) while conduct problems are linked to an underactive behavioral inhibition system (BIS). However, this research mostly uses adult samples, cross-sectional designs, and only tests between-person associations. Moreover, most studies typically only test one direction of effects (i.e., temperament predicts externalizing problems) and do not consider alternative models of associations, such as reciprocal associations. To address these limitations, we use three annual waves of a longitudinal, community-based sample of 387 early adolescents (mean Wave 1 age = 11.61, 55% female, 83% non-Hispanic White) to test reciprocal associations between BIS and BAS and oppositional and conduct problems. Latent growth curve models with structured residuals (LCM-SR) are used to test hypotheses and disaggregate between- and within-person associations. Evidence supports within-person reciprocal associations between BAS and oppositional problems and between BIS and combined conduct and oppositional problems. Results potentially inform developmental theories of temperament and externalizing problems linkages and interventions with adolescents who are engaging in oppositional problems and more severe conduct problems.

Introduction

Temperament encompasses individual differences in reactivity and self-regulation that have a biological a basis, are shaped by environmental experiences, and guide how individuals navigate various environmental contexts (Rothbart & Bates, 2006; Shiner et al., 2021)¹. Early adolescence marks a sensitive period of temperament development due to a number of neurobiological and psychosocial changes that, in turn, place adolescents at greater risk for developing externalizing problems (e.g., aggression, anti-social behavior) (Cicchetti & Rogosch, 2002; Durbin & Hicks, 2014; Kessler et al., 2007; Hill & Edmonds, 2017; Nivard et al., 2017; Powers & Casey, 2015). Externalizing problems in adolescence are associated with a number of important adjustment outcomes, such as academic achievement and substance use (Colder et al., 2018; Okano et al., 2020), highlighting the public health significance of better understanding how such problems develop. Prominent developmental theories and empirical evidence suggest associations between temperament and externalizing problems and thus support exploring temperament as a means to better understand risk for externalizing problems (Fowles & Dindo, 2009; Rothbart, 1989; Steinberg, 2008), but the nature of these associations remains equivocal. Gray’s Reinforcement Sensitivity Theory (RST) is a learning model of individual differences that conceptualizes temperament as being comprised of motivational systems that underlie approach and avoidance behaviors (Gray & McNaughton, 2000). It has been used to help explain externalizing problems in part because as a learning model it allows for integration of biologically-based individual differences into social learning theories (Bijttebier et al., 2009). For example, strong approach motivation may increase vulnerability for aggression in pursuing potential rewards or low avoidance motivation may lead to increased vulnerability for delinquency because of less sensitivity of punishment and hence low likelihood of inhibiting transgressive behavior. However, several limitations of this literature limit our understanding of how these constructs are related to one another and co-develop during adolescence. Specifically, the current study aims to address: 1) an underutilization of child and adolescent samples that could offer a developmental perspective on RST and externalizing problems associations, 2) limited longitudinal work that can test temporal precedence, co-development, and change over time at the within-person level, and 3) a lack of explicitly tested theoretical models of temperament and externalizing problems relations and bidirectional influences (Alloy et al., 2016; Durbin & Hicks, 2014; Tackett, 2006; Katz et al., 2020).

Temperament and Psychopathology Models: Considering Within-Person Effects

The most prominent model of testing RST and externalizing problems associations is the vulnerability model, which posits that temperament confers risk for psychopathology (e.g., De Bolle et al., 2012). Specifically, varying levels of RST dimensions confer risk for certain externalizing problems outcomes, like conduct problems or aggression (Bijttebier et al., 2009; Katz et al., 2020; Morgan et al., 2014). An alternative model, initially proposed by Lewinsohn and colleagues (1981), is the scar model. The scar model posits that experiencing intense psychopathology symptoms leaves a lasting impact on temperament, which could manifest as temporary (i.e., scab effect; Li & Zinbarg, 2007) or long-lasting (i.e., scar effect) changes in temperament. In other words, elevated externalizing problems predict changes in RST motivational systems. However, little scar effect work has been done in adolescence or with RST, and evidence has not always been consistent (Williams et al., 2021). Finally, some evidence suggests these two models operate simultaneously as reciprocal associations (Brotman et al., 2017; Durbin & Hicks, 2014; Katz et al., 2020; Mogg & Bradley, 2016, 2018; Pinto-Meza et al., 2006; Van Bockstaele et al., 2014). Reciprocal effects are tested less frequently, particularly with regard to RST constructs, and have historically been tested using statistical analysis models that conflate between and within-person associations (Hamaker et al., 2015).

Notably, these temperament and psychopathology models purport both between-person (i.e., rank order change between individuals) and within-person associations (i.e., time-specific variations of repeated measures within the same individual). For example, a vulnerability perspective could suggest that an adolescent with higher reward sensitivity relative to others may be at greater risk for externalizing problems relative to others (between-person). Simultaneously, if an individual shows higher levels of reward sensitivity at one time point than is typical for them, they may be at risk for a similarly larger than typical increase in externalizing problems at the next time point (within-person). Within-person analyses allow researchers to disaggregate between- and within-person associations of two variables while accounting for between-person mean levels and/or growth (Curran & Hancock, 2021; Curran et al., 2014; Hamaker et al., 2015). This can be accomplished with latent curve models with structure residuals (LCM-SR; Curran et al., 2014), which are used in the current study. This allowed us to test growth and within-person cross-lagged associations between RST and externalizing problem across early adolescence. Crossed-lagged associations provide a way to considering both vulnerability (temperament predicting externalizing problems) and scar (externalizing problems predicting temperament) in the same model. We acknowledge that there are other viable models that inform linkages between temperament and psychopathology (e.g., spectrum model, common cause model; Durbin & Hicks, 2014; Tackett, 2006). We selected the vulnerability, scar, and bidirectional models because we found them conceptually compelling. The proposed analysis and data structure (repeated annual waves) are well suited for testing these models. We review the predominantly between-person RST and externalizing literature to inform hypotheses about within-person vulnerability, scar, and reciprocal associations (Bijttebier et al., 2009; Katz et al., 2020).

Temperament – Reinforcement Sensitivity Theory (RST)

Gray’s revised Reinforcement Sensitivity Theory (RST) conceptualizes temperament as individual differences in sensitivity to rewarding or aversive stimuli (Gray & McNaughton, 2000). At its core, RST is a learning model in which individuals form expectations about their behavior and desirable, aversive, or uncertain outcomes. Although originally developed using animal models of behavior, RST has increasingly been applied to understand human behavior, affect, and temperament and is grounded strongly in neuroscience (Corr, 2002; Adrián-Ventura et al., 2019; Warr et al., 2020). RST conceptualizes temperament as comprised of three, interdependent motivational systems that guide human behavior (Corr, 2002, 2008; Gray & McNaughton, 2000): The Behavioral Activation System (BAS), Behavioral Inhibition System (BIS), and Fight-Flight-Freeze System (FFFS). The BAS governs reward prediction and approach tendencies towards potentially rewarding or appetitive stimuli (Pickering & Gray, 2001; Smillie, 2008). Because the BAS is heavily influenced by dopaminergic activity in reward system regions of the brain (Smillie, 2008), BAS and sensitivity to reward are often used synonymously. The FFFS system is hypothesized to mediate reactions to conditioned and unconditioned aversive stimuli and is strongly linked to fear responses, leading to avoidance or escape tendencies (Gray & McNaughton, 2000). The BIS system is conceptualized as a conflict resolution system, eliciting increased attention to threating stimuli and moderating the conflicting goals of simultaneous BAS and FFFS activation (Corr, 2002; Gray & McNaughton, 2000). The BIS is thought to be associated with worry and anxiety, and is implicated in negative mood (Katz et al., 2020). In other words, while the BAS mediates reward seeking behaviors, the BIS mediates reactions to aversive stimuli and is often referred to as sensitivity to punishment. Given that the FFFS and BIS are highly correlated and that the BIS and BAS systems are theoretically and empirically associated with a wide spectrum of externalizing problems, the current study focuses on the BIS and BAS systems (Bijttebier et al., 2009; Katz et al., 2020).

RST Development

Evidence for BIS and BAS is found in early childhood, as young as three or four years of age (Blair, 2003; Blair et al., 2004). The majority of BIS and BAS work in childhood and early adolescence has been cross-sectional, but this work still demonstrates that BIS and BAS motivational systems are associated with a number of important adjustment outcomes across development, including classroom behavior (Blair, 2003), personality development (Slobodskaya & Kuznetsova, 2013), and affect and substance use (Rádosi et al., 2021). Limited longitudinal work examining BAS development generally shows that BAS increases across adolescence, with a peak in mid- to late- adolescence, and evidence is mixed for linear versus non-linear growth (Colder et al., 2013; Gray et al., 2016; Schreuders et al., 2018; Urošević et al., 2012). Limited evidence also suggests that BIS shows normative decreases across early adolescence (Colder et al., 2013). These patterns are consistent with similar models of motivational aspects of temperament/personality development (e.g., Crawford et al., 2003; Gullo & Dawe, 2008; Klimstra et al., 2009; Steinberg, 2005). However, other studies have found no effect of age on levels of BIS/BAS (Braams et al., 2015) or have suggested that different facets of BAS may show differential developmental patterns (Pagliaccio et al., 2016; Schreuders et al., 2018; Urošević et al., 2012; Vervoort et al., 2015). The majority of this work has relied on between-group comparisons (e.g., BAS levels in children versus adolescents), which limits our understanding of mean changes and variability in individual trajectories of change in BIS and BAS over time. This also precludes testing the important question of whether individual deviations (within-person effects) from mean growth in BIS and BAS (between-person effects) meaningfully predict within-person deviations in externalizing problems. Furthermore, longitudinal designs are necessary to test mean growth trajectories (and individual variability in growth trajectories) in constructs, establish temporal precedence, and elucidate direction of effects between temperament and externalizing problems (Durbin & Hicks, 2014; Van Bockstaele et al., 2014).

RST and Externalizing Problems in Adolescence

Although prior work demonstrates the existence of a meaningful higher order externalizing problems factor (Forbes et al., 2016; Verona et al., 2011), externalizing problems have theoretically been distinguished into several subdomains. The two developmental phenotypes of interest in the current study include – oppositional problems, characterized by emotion dysregulation, anger and frustration, and reactivity to perceived threats; and conduct problems, characterized by low levels of fear, emotional and physiological hypoarousal, deficits in guilt and empathy, and proactive (i.e., goal-oriented) aggression (e.g., Frick & Morris, 2004; Wakschlag et al., 2018). In general, conduct problems tend to reflect a more severe, persistent course of externalizing problems relative to oppositional problems (Frick & Morris, 2004). Furthermore, whereas most individuals with oppositional problems do not demonstrate or develop these more severe conduct problems, the majority of individuals with conduct problems show comorbid oppositional problems (Maughan et al., 2004; Nock et al., 2007).

Relatively few studies have directly examined associations between RST and these dimensional facets of externalizing problems, or considered within-person effects. Instead, prior research has often focused on clinical cutoffs (i.e., categorical diagnoses), as captured by oppositional-defiant disorder (ODD) and conduct disorder (CD), or on constructs theorized to be closely related to the present study’s conceptualization of oppositional problems (e.g., anger, reactive aggression) and conduct problems (e.g., delinquency, callous-unemotional traits, proactive aggression). Evidence suggests that dimensional measures better capture the broad variability of adolescent externalizing behaviors (e.g., Walton et al., 2011), and the current study’s community sample is more likely to present with sub-clinical levels of externalizing problems. Therefore, we focus on dimensional scales of oppositional-defiant and conduct problems as opposed to clinical cutoffs for categorical diagnoses. However, given the dearth of literature focusing specifically on these dimensions and RST, we consider literature examining both clinical categorical diagnoses and closely related emotional and behavioral constructs in our review of the literature below. We also briefly consider work conducted in adult samples given the limited research conducted on RST and externalizing problems in adolescence.

BAS and Externalizing Problems

Differences in reward and punishment sensitivity, as captured in BAS and BIS, have been highlighted as transdiagnostic features of externalizing problems (Nusslock & Alloy, 2017; Zald & Treadway, 2017). BAS hypersensitivity has been consistently theorized to be associated with a tendency toward broad externalizing problems (Bijttebier et al., 2009; Hundt et al., 2008; Morgan et al., 2014). Individuals with higher levels of BAS sensitivity are more sensitive to cues to potential reward, have higher behavioral approach responses to those cues, and may be biased towards attending to reward cues (e.g., Bjørnebekk, 2007). This reward and approach bias often comes at the expense of attending to potential punishment cues or weighing consequences, which may result in increased vulnerability for aggressive means towards pursuing goals, frustration when rewards are not obtained, or unanticipated longer-term consequences. Past work suggests that BAS is positively related to a number constructs closely related to oppositional problems, including anger, reactive aggression (i.e., aggression enacted in response to perceived threat), and impulsivity (Bijttebier et al., 2009; Johnson et al., 2014). Although the majority of this work has been done in adult samples (Bijttebier et al., 2009), a growing literature has established similar findings in adolescent samples – elevated BAS is predictive of overall externalizing problems, impulsivity, and reactive aggression (Bjørnebekk, 2007; Colder & O’Connor, 2004; Harmon-Jones & Peterson, 2008; Matthys et al., 1998; Morgan et al., 2014; Muris et al., 2005; Vervoort et al., 2015), with a few exceptions (Hasking, 2007; Fite et al., 2019; Roose et al., 2011). These findings suggest that while BAS hypersensitivity is predictive of externalizing behaviors broadly, it may especially reflect the emotionally-driven, frustrative phenotype characteristic of oppositional problems.

BIS and Externalizing Problems

In contrast to BAS, BIS hyposensitivity may specifically predict emotionally underreactive aspects of externalizing problems characteristic of conduct problems (e.g., Bjørnebekk, 2007; Morgan et al., 2014; Johnson et al., 2014). Individuals with BIS hyposensitivity show lower sensitivity to punishment cues or weaker effects of punishment for themselves or others, which may also make it more difficult to socialize them against engaging in externalizing behaviors, consistent with the more severe, persistent course associated with conduct problems (Bjørnebekk, 2007). In adult samples, hypoactive BIS has been found to predict a more severe, interpersonally detached pattern of antisocial behavior, including characteristics of psychopathy (Bijttebier et al., 2009). Although fewer studies have examined associations between RST and conduct problems in child and adolescent samples, Morgan and colleagues (2014) found that adolescents convicted of crimes showed lower levels of BIS relative to non-offending adolescents. Furthermore, whereas BAS was positively associated with several dimensions of antisocial behavior (e.g., impulsivity, grandiosity, conduct problems), lower BIS specifically predicted greater levels of callousness, although BIS was unrelated to self-reported conduct problems (Morgan et al., 2014). Other work in adolescent samples also support the finding that a lower BIS may specifically be indicative of a number constructs closely related to the present study’s conceptualization of conduct problems, including callousness and proactive aggression (Bjørnebekk, 2007; Matthys et al., 1998; Roose et al., 2011), with one exception that found no relation between BIS and delinquency (Hasking, 2007). Another exception found that in an incarcerated sample proactive aggression was highest for individuals with high BIS and low levels of perceived containment (i.e., the perception that others can control their behavior; Fite et al., 2019). The authors suggest this indicates that hypersensitivity to threatening cues may prompt proactive aggression when individuals also believe others may not be able to control their behavior (Fite et al., 2019). Despite these exceptions, taken together evidence suggests that whereas a hyperactive BAS may be predictive of behaviors consistent with oppositional problems (e.g., impulsivity, anger, reactive aggression) or combined oppositional and conduct problems, a hypoactive BIS may more specifically predict the more severe, callous, and proactive externalizing behaviors reflective of conduct problems.

Current Study

Despite long-standing calls for empirical work examining potential scar or reciprocal effects of RST constructs and psychopathology, especially utilizing prospective longitudinal designs to test within-person associations (e.g., Bijttebier et al., 2009; Durbin & Hicks, 2014; Katz et al., 2020), very few studies have addressed these limitations in the literature. The present study aimed to address these gaps by using an early adolescent sample to examine longitudinal associations between BIS, BAS, and two externalizing domains: oppositional-defiant problems and conduct problems. Additionally, we employed a statistical model (LCM-SR; Curran et al., 2014) that disentangles change at both the between- and within-person levels, as well as potential reciprocal within-person associations, while also controlling for between-person levels and growth. In terms of growth across early adolescence, we hypothesized that 1) BAS, oppositional problems, and conduct problems would increase over time and 2) BIS levels would decline. At the within-person level, we expected reciprocal associations to be supported between BAS and oppositional-defiant problems and between BIS and conduct problems. Specifically, we hypothesized that 3) higher than expected BAS would predict elevated oppositional-defiant problems after accounting for growth (vulnerability effect) and oppositional-defiant problems, in turn, would predict elevated BAS levels above and beyond growth (scar effect), and 4) lower than expected levels of BIS would predict elevated conduct problems above and beyond growth and conduct problems, in turn, would predict lower BIS above and beyond growth.

Methods

Participants

The community sample was recruited in Erie County, NY, for longitudinal study on investigating substance use and psychopathology across adolescence. The sample was recruited to be representative of the local county with regard to SES and race and recruitment occurred December 2006 to February 2009. Eligibility criteria included being between 11–12 years old at the time of recruitment, speaking English fluently, and having no disabilities that would preclude completion of the assessment battery. In the current study, we utilized the first three annual waves of the larger longitudinal study when the variables of interest were assessed. The sample at Wave 1 included 387 families (1 adolescent and 1 parent/caregiver), 373 families at Wave 2 (4% attrition), and 370 families at Wave 3 (4% attrition). Adolescent mean age at Wave 1 was 11.61 (SD = .09), 55% of the sample were female, and the median family income was $70,000 (in the range of $1,500 to $500,000) and 6% of families received public income assistance. The majority of adolescents were non-Hispanic White (83%), followed by 9% Black, 4% other or mixed race, 2% Hispanic/Latino, 1% Asian American/Pacific Islander, and <1% Native American. The majority of adolescents came from two-parent/caregiver families (75.7%) and the majority of caregivers who participated in data collection were biological mothers (87%). Overall retention across all three Waves was strong, ranging from N = 321 (95%) to N = 331 (98%). Prior work in this sample found no significant differences between those with and without missing data based on demographic and substance use variables (Meisel & Colder, 2021). The low attrition and lack of significant differences suggest that missing data did not substantially impact study findings.

Procedures

At all three waves, adolescents and their caregivers completed an assessment battery at the university research offices. After providing consent/assent, adolescents and caregivers were escorted to separate rooms to complete measures privately. Of interest to this study, parents completed measures of adolescents’ demographics and temperament (BIS and BAS) and the adolescents completed measures related to psychopathology symptoms. Families were compensated $75, $85, and $125 for their participation across Waves 1 to 3, respectively, and adolescents earned a small prize ranging from $5 to $15. All procedures were approved by the university’s institutional review board (IRB).

Measures

Demographics

Parents reported on adolescent gender (0 = male and 1 = female) and years of age.

BIS and BAS

The Revised Sensitivity to Punishment and Sensitivity to Reward Questionnaire for Children (SPSRQ-C revised; Colder et al., 2011; Colder & O’Connor, 2004) is a parent-report questionnaire consisting of 33 items and was adapted from Torrubia and colleagues’ (2001) Sensitivity to Punishment and Sensitivity to Reward Questionnaire (SPSRQ) to assess BIS and BAS. Caregivers responded on a 5-point response scale (1 = strongly disagree to 5 = strongly agree). Items were averaged to create two higher order scales: BIS (also known as sensitivity to punishment) and BAS (also referred to as sensitivity to reward). Two items were deleted from the BAS scale due to poor psychometric performance (see Colder & O’Connor, 2004). Reliability was excellent for BAS (Cronbach’s α = .82 to .86) and BIS (Cronbach’s α = .87 to .88) across all three waves of assessment.

Oppositional Problems and Conduct Problems

The Youth Self-Report scale (YSR; Achenbach & Rescorla, 2001) was used to measure adolescent externalizing symptoms. Adolescents responded on a 3-point response scale (0 = not true, 1 = somewhat or sometimes true, and 2 = very true or often true). The DSM-oriented Oppositional Problems (5 items) and Conduct Problems (15 items) subscales were used due to specific hypotheses about how each construct differently relates to BIS and BAS. Items were averaged to create scales for each wave. Reliability for oppositional problems ranged from acceptable to good (Cronbach’s α = .63 to .72) and ranged from good to excellent for conduct problems (Cronbach’s α = .77 to .81). Oppositional and conduct problems scales were also averaged across all three waves of assessment to use as a control variable in each model (e.g., controlling for Wave 1–3 conduct problems in the model predicting oppositional problems, and vice versa). This allowed us to control for the other externalizing problems scale in each model without adding a large number of parameters that would be required if they were controlled for at each wave. Reliability was excellent for both oppositional problems (Cronbach’s α = .81) and conduct problems (Cronbach’s α = .82) when combined across waves.

Anxiety and Depressive Problems

Broadly, evidence suggests that BIS is positively associated with anxiety, BAS is negatively associated with depression (Katz et al., 2020), and that internalizing and externalizing symptoms are often positively associated (e.g., Cosgrove et al., 2011). Failing to statistically controlling for anxiety and depressive problems might result in the hypothesized unique associations between BIS and conduct problems and BAS and oppositional-defiant problems, respectively, canceling out. As such, we decided to statistically control for anxiety and depressive problems in our models and used the DSM-oriented Anxiety Problems and Depressive Problems scales of the Youth Self-Report form. Adolescents responded on a 3-point response scale (0 = not true to 2 = very true or often true). Items were averaged to create individual anxiety and depressive problems scales for each wave and Cronbach α’s ranged from .69 to .77 for anxiety problems and .70 to .76 for depressive problems. Each scale for Waves 1 to 3 were then averaged to create overall composites of anxiety and depressive problems to use as control variable in the final models. Reliability for these scales were excellent: Cronbach α =.81 for anxiety problems and .83 for depressive problems.

Analytic Strategy

Hypotheses were tested using longitudinal growth curve modeling with structured residuals (LCM-SR; Curran et al., 2014) estimated in Mplus Version 8.0 using maximum likelihood robust estimation (MLR; Muthén & Muthén, 1998 –2017). Because longitudinal data is inherently multilevel (i.e., includes both within and between-person levels of data), developmental models of change and co-development of constructs must account for both levels (Curran & Hancock, 2021; Hamaker et al., 2015). Although rarely modeled as such, both the vulnerability and scar hypotheses are models of individual change, suggesting the need for repeated measures at the within-person level (e.g., Durbin & Hicks, 2014). The LCM-SR (Curran et al., 2014) was used for the current study to account for developmental trajectories of psychopathology (e.g., Kessler, 2007) and BIS/BAS (Colder et al., 2013) and model within-person reciprocal associations. The LCM-SR maps onto the vulnerability and scar models because it allows for testing how deviations in typical trajectories of temperament and psychopathology (i.e., high or low levels above and beyond expected mean levels and growth) impact one another over time.

Following Curran and colleagues’ (2014) approach, models were constructed in a step-wise fashion. First, we estimated unconditional univariate growth models for each dimension of externalizing behavior (oppositional and conduct problems) and temperament (BIS and BAS). Due to only three time points of data available, we only tested intercept only and linear growth models. Slope factor loadings (0, 1, and 2) were set so that the intercept represented Wave 1 of each variable. Factor covariances and autoregressive paths between residual variances, including equality constraints among stability pathways, were tested using nested chi-square (χ²) difference tests and changes in CFI (Chen, 2007; Pavlov et al., 2020). Model fit was assessed using conventional absolute and incremental fit indices. Because cutoffs for “good” fit can differ between models, ranges were used to determine acceptability of model fit (Hu & Bentler, 1999; Marsh et al., 2004). Fit indices and ranges included model chi-square (a significant chi-square indicates poor fit), the comparative fit index (CFI) and Tucker-Lewis index (TLI; for both <.90 is poor, .90 to .94 is acceptable, and ≥.95 is excellent), root mean square error approximation (RMSEA; >.08 is poor, .05 to .07 is acceptable, and ≤.05 is excellent), and standardized root mean square residual (SRMR; >.09 is poor, .06 to .09 is acceptable, and ≤.06 is excellent). After establishing the best fit for each univariate growth model, they were combined into two multivariate LCM-SRs: an oppositional-problems and conduct problems model. We elected to run two separate multivariate models due to the complexity of LCM-SRs (e.g., sample size to model parameter ratio, only three repeated measures) and combing the two multivariate models would likely result in non-convergence. As a compromise to having all externalizing problems in the same model, we statistically controlled for externalizing symptoms averaged across all three waves that were not the focus of the primary model (i.e., Waves 1–3 oppositional problems controlled for in the model predicting conduct problems and Waves 1–3 conduct problems controlled for in the model predicting oppositional problems). Given the high comorbidity between opposition and conduct problems but distinct patterns of findings in how they are related to BIS and BAS (e.g., Nock et al., 2007), final models were tested both with and without this control variable in order to understand how the potential vulnerability and scar pathways operated for oppositional problems while controlling for overlapping variance with conduct problems, and vice versa. In order to test both the scar and vulnerability models (i.e., reciprocal effects), all hypothesized cross-lags between temperament and psychopathology residual variances were estimated and equality constraints across waves were tested. Psychopathology and temperament intercept and slope factors were allowed to covary and were regressed on statistical control variables, including gender and age, anxiety and depressive problems, and the alternative externalizing problems composite (see Figure 1 for conceptual figure). In order to reduce the impact of non-normality and missing data on results, conduct problems were windsorized at three standard deviations from the mean (see Table 1 for transformed skew and kurtosis) and robust maximum likelihood (MLR) estimation was used. The data and codebooks for this project are publicly available at Inter-university Consortium for Political and Social Research. The study analysis code will be sent to interested parties upon request. Study hypotheses and analyses were not pre-registered.

Figure 1.

Conceptual figure. Covariances were estimated between all exogenous predictors, cross-construct intercept and slope factors, and within-time cross-construct residuals but not depicted to simplify the figure. All paths of latent intercepts and slopes regressed onto observed control variables not show for visual interpretation. Identical model estimated where opposition problems and conduct problems are switched. BIS = Behavioral inhibition system, BAS = Behavioral activation system, CP = Conduct problems, OP = Oppositional problems, ANX = Anxiety problems, DEP = Depressive problems.

Table 1.

Bivariate Correlations and Descriptive Statistics

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16
Wave 1
1. Gender	-
2. Age	.05	-
3. BAS	−.17^	.01	-
4. BIS	.02	.04	.02	-
5. OP	−.14^	.09	.22	−.0002	-
6. CP	−.28	.05	.26	.04	.61	-
Wave 2
7. BAS	−.15^	.001	.79	−.0002	.20	.19	-
8. BIS	.10	.04	−.04	.81	−.02	.01	.01	-
9. OP	−.08	.03	.23	−.04	.57	.45	.19	−.06	-
10. CP	−.24	.02	.26	−.01	.50	.65	.19	−.02	.66	-
Wave 3
11. BAS	−.10*	.03	.76	−.02	.18	.21	.77	−.02	.22	.24	-
12. BIS	.13^	.07	.03	.74	.05	.04	−.0002	.79	−.02	.0003	.07	-
13. OP	.02	.09	.21	−.001	.52	.38	.18	−.01	.70	.56	.24	.04	-
14. CP	−.15	.06	.25	.02	.38	.49	.17	.01	.55	.69	.25	.05	.69	-
Waves 1–3
15. Anx	.02	.05	−.005	.12*	.36	.36	−.02	.16^	.44	.40	−.005	.16^	.44	.33	-
16. Dep	−.12*	.02	.16^	.08	.45	.51	.16^	.13*	.47	.56	.15^	.12*	.51	.52	.67	-
Mean	0.55	11.61	3.15	2.71	0.46	0.12	3.10	2.67	0.49	0.15	3.06	2.61	0.52	0.16	0.34	0.21
SD	0.50	0.55	0.46	0.53	0.34	0.15	0.44	0.54	0.37	0.16	0.48	0.50	0.41	0.18	0.27	0.20
Skew	−0.20	0.13	−0.01	0.37	0.55	1.59	0.11	0.48	0.51	1.29	0.06	0.53	0.54	1.31	0.88	1.47
Kurtosis	−1.97	−.092	0.36	−0.09	−0.15	2.40	0.13	0.53	−0.22	1.24	0.06	0.40	−0.30	1.33	0.44	2.17

Note. Winsorized CP are reported. BAS = Behavioral activation system, BIS = Behavioral inhibition system, OP = Oppositional problems, CP = Conduct problems, Anx = Anxiety problems, Dep = Depressive problems.

^*= p < .05,

^{^}= p < .01, Bolded = p <.001.

Results

Descriptive statistics and bivariate correlations of observed variables are reported in Table 1.

Univariate Growth Models

Information on univariate growth model construction is reported in Supplemental Materials 1.

BAS

The final univariate linear growth model for BAS fit the data well (χ² (2 df) = 0.82, p = .66; CFI = 1.00; TLI = 1.00; RMSEA = .00; SRMR = .04) and suggested significant declines in BAS (intercept M = 3.14, SE = 0.2, p < .001; slope M = −0.05, SE = 0.01, p < .001)². A random intercept (σ² = .16, SE = .01, p < .001) was supported but and random slope was not (p = .87), suggesting significant variability in adolescents’ Wave 1 BAS levels but not in rate of change over early adolescence.

BIS

The final univariate linear growth model for BIS fit the data well (χ² (1 df) = 0.40, p = .52; CFI = 1.00; TLI = 1.00; RMSEA = .00; SRMR = .01) and suggested significant declines in BIS (intercept M = 2.71, SE = 0.03, p < .001; slope M = −0.05, SE = 0.01, p < .001). A random intercept (σ² = .26, SE = .03, p < .001) and random slope were supported (σ² = .03, SE = .01, p = .001). This suggests significant variability in adolescents’ Wave 1 BIS levels and in their rates of change over time.

Oppositional Problems

The final univariate linear growth model for oppositional problems fit the data well (χ² (2 df) = 0.62, p = .97; CFI = 1.00; TLI = 1.00; RMSEA = .00; SRMR = .002) and suggested significant increases in oppositional problems (intercept M = 0.46, SE = 0.02, p < .001; slope M = 0.03, SE = 0.01, p < .001). A random intercept (σ² = .07, SE = .01, p < .001) and random slope were supported (σ² = .02, SE = .003, p < .001), suggesting significant variability in adolescents’ Wave 1 oppositional problems and in rate of change over early adolescence.

Conduct Problems

The final univariate linear growth model for conduct problems fit the data well (χ² (2 df) = 3.17, p = .21; CFI = .99; TLI = .99; RMSEA = .04; SRMR = .04) and suggested significant increases in conduct problems (intercept M = 0.13, SE = 0.01, p < .001; slope M = 0.02, SE = 0.004, p < .001). A random intercept (σ² = .02, SE = .002, p < .001) and random slope were supported (σ² = .003, SE = .001, p < .001), suggesting significant variability in adolescents’ Wave 1 conduct problems and in rate of change over early adolescence.

Structural Models – LCM-SR

Details for each model can be found in Supplemental Materials 1. Of note, the variance for the latent BAS slope factor was set to 0 in all structural models because it was not statistically significant in the univariate growth model and estimating it in each full LCM-SR model produced an error stating that a covariance matrix was not positive definite.

Oppositional Problems

The oppositional problems LCM-SR with conduct problems as a statistical control variable provided excellent fit to the data (χ² (38 df) = 38.37, p = .45; CFI = 1.00; TLI = 1.00; RMSEA = .01; SRMR = .02). Between-person associations for the oppositional problems LCM-SR are reported in Table 2. The BAS latent intercept was positively associated with the BIS latent intercept, indicating that children high in BAS at Wave 1 were also high in BIS. The oppositional problems and BAS latent intercept were positively and negatively associated with the Wave 1–3 anxiety problems composite, respectively. The oppositional problems latent intercept was also positively associated with the conduct problems Wave 1–3 composite. The BIS latent slope was negatively associated with the BIS latent intercept and gender, such that high initial levels of BIS were associated greater declines in BIS and girls exhibited less steep declines in BIS. Finally, the oppositional problems latent slope was positively associated with the conduct problems Wave 1–3 composite and gender, such that girls exhibited steeper growth in oppositional problems.

Table 2.

Between-Person Associations for LCM-SRs

	Oppositional Problems					Conduct Problems
Variable	1	2	3	4	5	1	2	3	4	5
1. BAS Int.	-					-
2. BIS Int.	.001	-				.01	-
3. BIS Slope	.07	−.42 ^	-			.06	−.42 ^	-
4. OP Int./CP Int.	.07	−.04	−.003	-		.14	−.07	.18	-
5. OP Slope/ CP Slope	.03	−.10	.14	-	-	−.07	.06	−.23	-	-
6. T1–3 CP/OP	.27+	−.07	.04	.69+	.34^	.28+	−.13*	.03	.52+	.48^
7. T1–3 ANX	−.23^	.12	−.02	.21^	.06	−.28+	.14	−.02	.03	−.31
8. T1–3 DEP	.15	.05	.08	.07	.07	.19*	.06	.09	.34+	.21
9. Gender	−.07	.01	.19^	−.02	.32^	−.11*	.02	.18^	−.29+	.18
10. Age	.01	.03	.01	.05	.003	.00	.04	.02	.01	.003

Note. Bolded = correlations, non-bolded = standardized regression coefficients. Statistics reported for models controlling for other externalizing problems. Gender was coded such that 0 = male and 1 = female. BAS = Behavioral activation system, BIS = Behavioral inhibition system, OP = Oppositional problems, CP = Conduct problems, Anx = Anxiety problems, Dep = Depressive problems.

^*= p < .05,

^{^}= p < .01,

⁺= p <.001.

Both the vulnerability and scar hypotheses were tested simultaneously by evaluating the within-person cross-lagged structural pathways. Within-person effects for oppositional problems models are reported in Figure 2. BAS residuals significantly and positively predicted oppositional problems residuals (βs = .16, SEs = .07, ps = .03) supporting our vulnerability hypothesis. This suggests that higher than expected levels of BAS (given growth) were prospectively associated with elevated levels of oppositional problem symptoms. The scar hypothesis was also supported, such that oppositional problems residuals significantly and positively predicted BAS residuals (Wave 1 predicting 2: β = .19, SE = .10, p = .04; Wave 2 predicting 3: β = .14, SE = .07, p = .04). That is, higher than expected levels of oppositional symptoms given growth were prospectively associated with high levels of BAS. Neither vulnerability effects (ps = .26) nor scar effects (ps = .16) were supported between BIS and oppositional problems. In sum, evidence supports reciprocal associations such that deviations from developmentally typical levels of BAS and oppositional problems confer risk for greater increases in one another.

Figure 2.

Within-person associations depicted. Hypothesized reciprocal associations depicted as solid lines; Standardized estimates, with standard errors reporting in parentheses, reported to the left of structural pathway. Controlling for other externalizing subfactor/Not controlling for other externalizing subfactor *= p < .05, ^ = p < .01, Bolded = p <.001.

The oppositional problems model without conduct problems as a statistical control variable also provided excellent fit to the data (χ² (34 df) = 28.46, p = .74; CFI = 1.00; TLI = 1.00; RMSEA = .00; SRMR = .02). Interestingly, this resulted in changes in the cross-lag paths. The vulnerability cross-lags became only marginally significant (Wave 1 predicting 2: β = .16, SE = .08, p = .05; Wave 2 predicting 3: β = .17, SE = .08, p = .05) whereas the scar pathways became non-significant (ps = .21). Taken together, this pattern of results from the two models provides tentative evidence that within-person reciprocal associations are supported between symptoms unique to oppositional problems (i.e., when variance from conduct problems is controlled for) and BAS instead of overlapping symptoms (i.e., conduct problems covariance not controlled for) and BAS.

Conduct Problems

The conduct problems LCM-SR with oppositional problems controlled for was tested first and provided excellent fit to the data (χ² (38 df) = 48.01, p = .13; CFI = 1.00; TLI = .99; RMSEA = .03; SRMR = .03). See Supplemental Materials 1 for more details on model construction. Between-person results were largely similar to the oppositional problems model (see Table 2 for between-person associations in the model with controlling for Wave 1–3 oppositional problems). The differences that emerged were that the BAS latent intercept was also positively associated with the oppositional problems and depressive problems Wave 1–3 composites and negatively associated with gender, such that those with elevated oppositional and depressive problems across Waves 1–3 and girls exhibited higher starting levels of BAS. Other differences between the two models suggested that the conduct problems latent intercept was negatively associated with the depressive problems Wave 1–3 composite, suggesting those with high starting levels of conduct problems at Wave 1 had low depressive problems, and that the conduct problems latent slope was not significantly associated with gender. The BIS latent slope was also negatively associated with Wave 1–3 oppositional problems indicating that high levels of oppositional problems were associated with steep declines in BIS.

Within-person effects for both conduct problems models are reported in Figure 2. Contrary to hypotheses, neither vulnerability effects (ps = .52) nor scar effects (ps = .87) were supported between conduct problems and BIS. The same was true for vulnerability (ps = .74) and scar (ps = .88) pathways between conduct problems and BAS. Next, the conduct problems model without oppositional problems controlled for was estimated and also provided excellent fit to the data (χ² (34 df) = 36.26, p = .36; CFI = 1.00; TLI = 1.00; RMSEA = .01; SRMR = .03). The pattern of results changed when oppositional problems were excluded from the model. Reciprocal associations were supported between conduct problems and BIS residuals, such that BIS residuals significantly and negatively predicted conduct problems residuals (Wave 1 predicting 2: β = −.35, SE = .19, p = .046 .05; Wave 2 predicting 3: β = −.42, SE = .20, p = .046) and, in turn, conduct problems residuals negatively predicted BIS residuals (Wave 1 predicting 2: β = −.32, SE = .14, p = .03; Wave 2 predicting 3: β = −.39, SE = .16, p = .03). These results suggest that deviations of BIS and conduct problems prospectively influence once another, such that, after accounting for mean growth, elevated conduct problems at one wave are associated with low BIS at the next wave and, in turn, low BIS predicts high conduct problems. The pattern of results from the two conduct problems models provides tentative evidence that within-person reciprocal associations are supported between conduct problems and BIS but only when the variance from oppositional problems is not controlled for (i.e., combined presentation of oppositional and conduct problems).

Broad Externalizing Problems

Based on the pattern of results from the oppositional and conduct models changing depending on whether the other externalizing problem subscale was included as a statistical control variable, a third, post-hoc model of broad externalizing problems was estimated. The broad externalizing symptoms scale combined conduct and oppositional problem symptoms into a single scale and yielded excellent reliability across all three waves (Cronbach’s α = .82 to .86). This model was constructed in the same fashion as the other two LCM-SRs. More information on unconditional univariate growth and full LCM-SR model construction as well as between-person associations can be found in Supplemental Materials 1. The final model provided excellent fit to the data (χ² (34 df) = 36.94, p = .33; CFI = 1.00; TLI = 1.00; RMSEA = .02; SRMR = .03). Similar to the conduct problems model without controlling for oppositional problems, reciprocal associations were supported such that BIS residuals negatively predicted externalizing problems residuals (Wave 1 predicting 2: β = −.37, SE = .15, p = .01, R² = .14, p = .18; Wave 2 predicting 3: β = −.50, SE = .17, p < .01, R² = .25, p = .12) and, in turn, externalizing problems residuals negatively predicted BIS residuals (Wave 1 predicting 2: β = −.37, SE = .11, p = .004, R² = .14, p = .10; Wave 2 predicting 3: β = −.47, SE = .14, p = .004, R² = .22, p = .10). This suggests that low BIS predicted elevated externalizing problems above and beyond mean change, which, in turn, predicted low BIS. Only vulnerability effects were supported between BAS and externalizing problems, such that elevated BAS predicted high externalizing problems above and beyond growth (βs = .14, SEs = .07, ps = .04). In summary, the broad externalizing model largely supported the findings from the individual oppositional and conduct problems models.

Discussion

The current study advances the RST and externalizing literature in a number of ways, including testing competing models of associations between BIS and BAS and oppositional and conduct problems (i.e., vulnerability, scar, and reciprocal effects), distinguishing between-person and within-person effects, and using longitudinal data to test temporal precedence in a sample of early adolescents, a sensitive period for temperament and psychopathology development (e.g., Durbin & Hicks, 2014). Evidence suggested that BIS and BAS declined across early adolescence while oppositional and conduct problems increased. We also found evidence for reciprocal associations between oppositional problems and BAS and conduct problems and BIS at the within-person level. However, these associations depended on whether the model was testing the unique effects of externalizing subscales (i.e., controlling for the other externalizing cluster of symptoms in the model) or broad externalizing symptoms. These findings suggest that RST motivational systems and externalizing problems co-develop over time at the within-person level and demonstrate the importance of considering both broad and specific sub-facets of externalizing problems when examining temperament and psychopathology links.

Between-Person Growth

The finding that BIS declined is consistent with prior work in this sample (Colder et al., 2013) that measured RST motivational systems via behavioral computer task. This finding is also consistent with other developmental models of temperament and personality that posit decreased motivational aspects of temperament related to weighing risks relative to reward (e.g., Steinberg, 2008). Increasing externalizing problems in early adolescence is also consistent with prior work (e.g., Galambos et al., 2003; Olson et al., 2013). During this period, adolescents engage in greater risk taking, become increasingly independent from and renegotiate relationship dynamics with parents and caregivers, and are more likely to associate with peers who engage in similar risk-taking or delinquent behavior (e.g., Hoeben et al., 2016; Rothbart, 1989).

Biological changes occurring during adolescence are also relevant to externalizing problems and BIS/BAS development, including pubertal and brain development, particularly in regions associated with reward processing (Steinberg, 2005). As such, we expected to observe increases in BAS, but found that BAS declined over the three annual assessments. This is in contrast to some prior work that utilized repeated measures during early adolescence and found evidence for increases in BAS via behavioral computer task (Colder et al., 2013) or no change (Braams et al., 2015). Post-hoc analyses suggested that the reward responsivity subscale declined across early adolescence, while the drive and impulsivity/fun seeking subscales did not significantly change (although the mean slope as also negative). This suggests that the reward responsiveness scale may show stronger declines and may underlie overall univariate BAS decline. This is consistent with other work that suggests that change may differ among the BAS subscales, however this work is mixed with regard to the pattern of change among subscales (Pagliaccio et al., 2016; Urošević et al., 2012). Differences in measurement (i.e., self-report, parent-report, and behavioral or computer tasks; Colder et al., 2013; Schreuders et al., 2018), varying time intervals, such as modeling cross-lags or age groups that include later childhood or middle to late adolescence (Vervoot et al., 2015), and relying on between-group analysis to test age differences (Pagliaccio et al., 2016; Urošević et al., 2012) limit our ability to place these findings in the context of the larger literature. The developmental trajectory of BAS during early adolescence remains an open question due to these mixed findings and limitations in modeling and research design. More work that addresses these limitations is needed to confirm the current findings of declines in BAS during early adolescence.

Within-Person Reciprocal Associations

Oppositional Problems

The oppositional problems models suggest that BAS and oppositional problems are reciprocally related, supporting both vulnerability and scar effects. In other words, when individuals showed higher than their typical levels of BAS, they engaged in more oppositional problems than would be expected one year later, and demonstrating greater than typical levels of oppositional problems predicted higher than expected levels of BAS one year later. Notably, these reciprocal associations were only supported in the model statistically controlling for conduct problems; when conduct problems were dropped as a control variable, the vulnerability effects became marginal and the scar effects were no longer significant. Additionally, the vulnerability effects were also supported in the post-hoc broad externalizing model, although scar effects remained non-significant. Together, this suggests that BAS vulnerability effects may apply to broad externalizing problems, consistent with prior work (Bijttebier et al., 2009; Hundt et al., 2008; Morgan et al., 2014), whereas reciprocal associations (i.e., vulnerability and scar effects) may be unique to oppositional problems that do not co-occur with conduct problems.

To our knowledge, this represents the first evidence of scar effects between BAS and oppositional problems, and suggests that reward sensitivity not only potentiates oppositional problems, but that oppositional problems may also lead individuals to develop a bias toward rewarding stimuli. The novelty of this finding and the small effects makes it difficult to place in context and draw firm conclusions. However, it is broadly consistent with pathophysiological models of irritability (Brotman et al., 2017), a key facet of oppositional problems (Evans et al., 2020), and of models of emotionally reactive, frustrative phenotypes of externalizing problems (Wakschlag et al., 2018), both of which posit sensitivity to reward receipt and omission as a central component of these constructs, suggesting they may develop in concert. Furthermore, that reciprocal effects emerged only when conduct problems were controlled suggests this process is specific to oppositional behaviors. Indeed, conduct problems may show a different pattern (i.e., hyposensitivity; Matthys et al., 2013) or no clear association (Wakschlag et al., 2018) with reward sensitivity, which could mask effects if not accounted for. However, these reciprocal effects await replication.

No significant vulnerability or scar effects were found between BIS and oppositional problems, with or without controlling for conduct problems. This is consistent with hypotheses and prior work suggesting that oppositional problems are primarily associated with strong approach tendencies (i.e., hyperactive BAS; Bjørnebekk, 2007; Colder & O’Connor, 2004; Harmon-Jones & Peterson, 2008), whereas hypoactive BIS may be more specifically associated with a more severe and proactive form of externalizing problems (e.g., Morgan et al., 2014). These assertions are further supported by our findings from the conduct problems models.

Conduct Problems

Within-person effects from conduct problems models were largely consistent with hypotheses, although this also depended on whether or not co-occurring oppositional problems were statistically controlled. Specifically, reciprocal negative associations were found between conduct problems and BIS in the model not controlling for co-occurring oppositional problems. These effects were also seen in the post-hoc broad externalizing model, suggesting that they are reflective of processes involving combined oppositional and conduct problems, and more severe externalizing problems overall. As conduct problems rarely emerge without co-occurring oppositional problems (Maughan et al., 2004; Nock et al., 2007) this may be more representative of the common presentation of conduct problems. The vulnerability effects are consistent with findings and theory that low levels of punishment sensitivity make it difficult for families and communities to socialize youth to conform to behavioral norms (e.g., Bjørnebekk, 2007; Dadds & Salmon, 2003), and that youth with low fear conditioning responses are more likely to develop callousness and antisocial behavior reflective of conduct problems (Chen et al., 2021; Gao et al., 2010). The observed scar effects are novel and small in magnitude, making it difficult to compare to prior work and draw firm conclusions. Nonetheless, the combined vulnerability and scar effects (i.e., reciprocal associations) provide some evidence to support developmental theories of more severe externalizing problems onset and development: adolescents who engage in elevated conduct and oppositional problems likely experience more punishment from their environment (e.g., getting in trouble at school or conflict with parents) and over time habituate to the repeated negative feedback they receive, which in turn puts them at greater risk for future problem behaviors (e.g., Dodge & Pettit, 2003; Patterson et al., 2017).

Within-person associations between BAS and conduct problems were not supported. Null findings were consistent regardless of whether oppositional problems were controlled and also in the post-hoc broad externalizing model. These findings are consistent with theory such that a hypoactive BIS is more closely associated with severe externalizing problems relative to BAS (e.g., Bijttebier et al., 2009). However, prior work provides some support for a positive association between BAS and characteristics of broad externalizing problems (i.e., vulnerability effects) in adults and adolescents (e.g., Harmon-Jones & Peterson, 2008; Hundt et al., 2008; Morgan et al., 2014). This work exclusively tests between-person associations that are likely confounded with within-person associations and is largely cross-sectional, which could explain differences found in the current models. Moreover, we found a positive association between the BAS latent intercept and the conduct problems latent intercept (when not controlling for oppositional problems) and vulnerability within-person effects in the broad externalizing model. Our results, along with this prior work, suggests that BAS may operate as a between-person risk factor for elevated combined conduct and oppositional problems, but a hypoactive BIS may drive individual change in combined externalizing problems and set adolescents on the riskiest developmental trajectory for more severe externalizing problems.

Limitations and Conclusions

The current results should be interpreted in light of study limitations. First, many observed longitudinal effects were small in magnitude. This is perhaps not surprising given that our models disaggregated associations into within and between-person components, and we agree with Adachi and Willoughby (2015) that small longitudinal effects can be important, particularly for theory development. Relatedly, this is the first application of LCM-SR models for testing bi-directional associations between temperament and externalizing problems, and findings require replication before drawing firm conclusions. Second, this is a community-based sample with non-clinical levels of oppositional and conduct problems. Future research should seek to replicate these longitudinal and within-person findings in clinical samples where externalizing problems are more prevalent. Third, our sample was not very diverse and it was small given the complexity of our models. Replicating and extending our work in more racially and ethnically diverse samples is an important future direction. Similarly, it was not feasible for us to examine gender or pubertal development differences. Potential gender and pubertal timing differences could be important since prior work both are associated with externalizing problems (Askari et al., 2022; Dimler et al., 2015) and this is another important direction for future research. Fourth, the current sample was recruited from 2006 to 2009, with data collection spanning 2006 to 2012 for all three waves of data. Askari and colleagues (2022) recently found evidence for increases in internalizing problems and decreases in externalizing problems among adolescents spanning from 2012 to 2017. These patterns suggest the current findings may not generalize to the experiences of present-day adolescents. Fifth, three repeated measures limited growth modeling to intercept-only and linear growth and future work should endeavor to include additional repeated measures to test non-linear and quadric growth and improve the performance of complex LCM-SRs (Curran et al., 2014). Sixth, given our sample size and the complexity of the models, we were unable to consider other variables that might further inform how BIS and BAS and externalizing problems co-development. Other individual-level factors, such as genetics or other aspects of temperament, may play an important role. For example, RST is considered a bottom-up or reactive conceptualization of temperament. Other top-down or regulation facets of temperament, like effortful control, likely impact whether BIS and BAS confer risk for psychopathology (Collins et al., 2017; Derryberry & Rothbart, 1997; Santens et al., 2020). Context variables could also provide greater insight into protective or risk factors for the observed effects. For example, parents and peers are robust influences on adolescent externalizing problems, and likely influence the co-development of temperament and externalizing problems (Patterson et al., 2017; Racz & McMahon, 2011). Additionally, other conceptual models, such as the common cause or spectrum models, may offer interesting insight into how temperament and psychopathology are related and co-develop over time (Durbin & Hicks, 2014; Tackett, 2006). Future work should seek to examine whether these other models are supported and how they compare to the models tested in the current study.

Despite these limitations, the current study has notable strengths, including explicitly testing models of temperament and psychopathology linkages against one another, disaggregating between- and within-between person effects as well as subcomponents of externalizing problems, and leveraging longitudinal data in a community sample spanning early adolescence in a latent growth modeling framework. Notably, temperament and externalizing problems associations were supported across reporters, increasing confidence in the robustness of the observed effects.

These findings also have interesting clinical implications that may warrant further investigation. The observed reciprocal effects, particularly between BIS and broad externalizing problems, provide some insight into the more persistent and severe presentations of combined conduct and oppositional problems (Frick & Morris, 2004). Consistent with prior research, our work suggests that assessing BIS may be useful diagnostically for identifying those at risk for early onset and more severe increases in externalizing problems (Bijttebier et al., 2009; Bjørnebekk, 2007). Relatedly, adolescents with low BIS levels may be less sensitive to punishment cues that can interrupt reward-driven or goal-directed behavior. This could partially explain why children and adolescents exhibiting elevated conduct problems, and thus low BIS, may be more responsive to treatments that de-emphasize punishment in favor of emphasizing reward, although these treatment effects have been inconsistent (Waschbusch et al., 2020; Hawes et al., 2014). It may be useful for future intervention work to test whether BIS is useful for identifying adolescents for whom this approach may be more effective, explore treatment mechanisms that might increase BIS sensitivity, or modify techniques to rely more on adolescent strengths (e.g., Sommers-Flanagan & Sommers-Flanagan, 1995).

Public Health Significance Statement:

This study found that early adolescents who were more sensitive to positive feedback (reward) were more likely to report aggressive behaviors and adolescents who were less sensitive to negative feedback (punishment) were more likely to report aggressive and rule breaking behavior and vice versa (e.g., more rule breaking and aggression leads to less sensitivity to negative feedback). These findings might help parents, therapists, and schools better manage problem behaviors.

Data Statement

The data and codebooks for this project are publicly available at Inter-university Consortium for Political and Social Research. The study analysis code will be sent to interested parties upon request. This study was not pre-registered.