Introduction
Executive functioning refers to a hypothesized family of top-down mental processes that support concentration, inhibition, attention, and behavioral regulation (Diamond, Reference Diamond2013). Healthy development of executive functioning is essential for mental and physical health, academic achievement, and success in life (Moffitt et al., Reference Moffitt, Arseneault, Belsky, Dickson, Hancox, Harrington, Houts, Poulton, Roberts, Ross, Sears, Thomson and Caspi2011; Royall et al., Reference Royall, Lauterbach, Cummings, Reeve, Rummans, Kaufer, LaFrance and Coffey2002). While executive functioning may be biologically based, it is also influenced by environmental experiences (Müller & Kerns, Reference Müller, Kerns, Liben, Müller and Lerner2015). It has been proposed that early adolescent alcohol use produces a cascade of physiological and neurobiological events that adversely affect trajectories of brain development (Crews et al., Reference Crews, He and Hodge2007), including the development of executive functioning (Squeglia et al., Reference Squeglia, Jacobus and Tapert2014). Despite some support for this idea (Wetherill et al., Reference Wetherill, Squeglia, Yang and Tapert2013; Thoma et al., Reference Thoma, Monnig, Lysne, Ruhl, Pommy, Bogenschutz, Tonigan and Yeo2011), other studies report no association between early alcohol use and subsequent executive functioning (Squeglia et al., Reference Squeglia, Spadoni, Infante, Myers and Tapert2009; Tapert et al., Reference Tapert, Granholm, Leedy and Brown2002). Indeed, in a review of this literature, Spear (Reference Spear2018) concluded that many of the studies that report negative neurocognitive consequences of adolescent alcohol use may overstate this conclusion; while there may be negative associations between alcohol use and performance on one or two cognitive tasks, alcohol-using youth compared to non-drinking controls did not exhibit altered performances on the majority of cognitive tasks examined. A better understanding of the ways in which early alcohol use may impact executive functioning development is of critical public health importance, and the current study aims to address this gap.
The extant literature that has investigated the association between adolescent alcohol use and subsequent executive functioning has two notable limitations that may explain the mixed empirical evidence. First, most studies have not considered alcohol use in the broader context of problem behavior (Squeglia et al., Reference Squeglia, Spadoni, Infante, Myers and Tapert2009; Wetherill et al., Reference Wetherill, Squeglia, Yang and Tapert2013). This leaves open the question of whether poor executive functioning may be due to the neurotoxic effects of alcohol or drug use, or a complex cascade involving externalizing symptoms and poor social adaptation. There is empirical evidence that the broader domain of externalizing symptoms increases risk for poor social adaptation which, in turn, hampers healthy executive functioning development (Eisenberg et al., Reference Eisenberg, Taylor, Widaman and Spinrad2015). Accordingly, it is important to consider the developmental context of adolescent alcohol use, and the potential pathways from externalizing problems, including consideration of indirect effects through social adjustment across several domains when examining associations between alcohol use and executive functioning. Failure to do so may yield a confusing and mixed pattern of results.
Second, the executive functioning literature faces significant challenges related to definition and measurement (Nigg, Reference Nigg2017). Indeed, the dominant framework posits that executive functioning (also commonly referred to as “self-regulation,” or “self-control”) is a broad construct comprised of a set of complex, modular functions (Miyake et al., Reference Miyake, Friedman, Emerson, Witzki, Howerter and Wager2000). However, this framework has led to considerable heterogeneity in measurement and terminology (Nigg, Reference Nigg2017) and has not been well-supported by subsequent psychometric studies (Karr et al., Reference Karr, Areshenkoff, Rast, Hofer, Iverson and Garcia-Barrera2018). This has likely contributed to confusion regarding the link between adolescent alcohol use and cognitive development. An alternative framework that has emerged from the computational modeling literature posits that individuals’ efficiency of accumulating goal-relevant evidence is a cognitive individual difference that underlies many of the modular executive functions (Löffler et al., Reference Löffler, Frischkorn, Hagemann, Sadus and Schubert2024; Weigard et al., Reference Weigard, Clark and Sripada2021; Weigard & Sripada, Reference Weigard and Sripada2021). This alternative framework shows great promise in unifying executive functions and has not been considered with respect to how alcohol may impact executive functioning.
To address these aforementioned gaps, the current study proposes to: (1) Utilize bifactor modeling to parse general externalizing symptoms and the unique, neurotoxic effects of alcohol use and other drug use, (2) Examine indirect effects to better understand how adolescent alcohol use, other drug use, the broader constellation of externalizing behavior, and social adaptation may be associated with the development of executive functioning, and (3) Apply diffusion decision modeling, a cognitive modeling approach, to behavioral data in order to provide a precise index of task-general executive functioning (efficiency of evidence accumulation; drift rate).
The externalizing context and social development
Externalizing symptoms have received little attention in studies examining the alcohol use-executive functioning link (Squeglia et al., Reference Squeglia, Spadoni, Infante, Myers and Tapert2009; Wetherill et al., Reference Wetherill, Squeglia, Yang and Tapert2013). The omission of externalizing symptoms is notable because past work considering the alcohol use-executive functioning association posits that the specific neurotoxic effects of alcohol use derail healthy executive functioning development (López-Caneda et al., Reference López-Caneda, Rodríguez Holguín, Cadaveira, Corral and Doallo2013); however, an alternative possibility is that, separate from alcohol use but highly comorbid, externalizing symptoms (e.g., aggressive and rule-breaking behaviors) increase risk for poor social adaption, which in turn, hampers executive functioning development. Externalizing symptoms and alcohol use (and perhaps other drug use) may each lead to poor executive functioning, but through different mechanisms (equifinality).
Several threads of research support the idea of executive functioning operating through social adaptation. The development of executive functioning is facilitated by strong interpersonal relationships and experiences that help scaffold development of these cognitive abilities. Indeed, there is empirical evidence that social factors, such as parental scaffolding and social modeling, facilitate executive functioning development (Carlson, Reference Carlson2003; Geeraerts et al., Reference Geeraerts, Endendijk, Deković, Huijding, Deater-Deckard and Mesman2021). This is important because externalizing symptoms may disrupt critical social interactions that promote executive functioning. Indeed, externalizing symptoms negatively impact relationships with peers and parents, and school engagement (Baumeister et al., Reference Baumeister, DeWall, Ciarocco and Twenge2005; Buist et al., Reference Buist, Deković, Meeus and van Aken2004; Moilanen et al., Reference Moilanen, Shaw and Maxwell2010). Poor social adaptation and low engagement in school, in turn, may hamper healthy development of executive functioning (Eisenberg et al., Reference Eisenberg, Taylor, Widaman and Spinrad2015; Farley & Kim-Spoon, Reference Farley and Kim-Spoon2014; Heylen et al., Reference Heylen, Vasey, Dujardin, Vandevivere, Braet, De Raedt and Bosmans2017, Reference Heylen, De Raedt, Verbruggen and Bosmans2019). Hence, failure to consider externalizing symptoms when testing associations between alcohol use and executive functioning leaves open the possibility that the relationship between alcohol use and poor executive functioning is due to externalizing symptoms rather than the neurotoxic effects of alcohol use. Consideration of the broader externalizing context is consistent with recent calls for developmental psychopathology research to account for the co-occurrence of related symptoms and the hierarchical structure of psychopathology (Colder et al., Reference Colder, Scalco, Trucco, Read, Lengua, Wieczorek and Hawk2013; Forbes et al., Reference Forbes, Tackett, Markon and Krueger2016; Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby, Brown, Carpenter, Caspi, Clark, Eaton, Forbes, Forbush, Goldberg, Hasin, Hyman, Ivanova, Lynam, Markon and Zimmerman2017). Yet, no studies, to our knowledge, have tested mechanisms of the association between externalizing symptoms and poor executive functioning in adolescence, and this remains a significant limitation in the literature. The current study addresses this gap.
Measurement of executive functioning
The majority of studies examining links between alcohol use and executive functioning have used assessments designed to measure distinct executive functions (e.g., attentional shifting, working memory, inhibitory control, etc.; Nigg, Reference Nigg2017), which often suffer from psychometric issues (Eisenberg et al., Reference Eisenberg, Bissett, Zeynep Enkavi, Li, MacKinnon, Marsch and Poldrack2019; Enkavi et al., Reference Enkavi, Eisenberg, Bissett, Mazza, MacKinnon, Marsch and Poldrack2019; Hedge et al., Reference Hedge, Powell and Sumner2018). For example, the Stop Signal Task (SST) measures the ability to suppress or stop prepotent responses inhibitory control, a key executive function (Diamond, Reference Diamond2013; López-Caneda et al., Reference López-Caneda, Rodríguez Holguín, Cadaveira, Corral and Doallo2013). Most studies using the SST and similar tasks rely on reaction time-based indices to measure inhibitory control. However, there is growing evidence that measures of Stop Signal Reaction Time (SSRT), response time (RT), and accuracy-based indices display poor to moderate test-retest reliability over short periods of time (e.g., three weeks; Hedge et al., Reference Hedge, Powell and Sumner2018; Thunberg et al., Reference Thunberg, Wiker, Bundt and Huster2023). Additionally, as noted above, fractionated approaches to conceptualizing and measuring executive functioning have demonstrated poor construct validity as individual difference dimensions (Karr et al., Reference Karr, Areshenkoff, Rast, Hofer, Iverson and Garcia-Barrera2018; Löffler et al., Reference Löffler, Frischkorn, Hagemann, Sadus and Schubert2024; Eisenberg et al., Reference Eisenberg, Bissett, Zeynep Enkavi, Li, MacKinnon, Marsch and Poldrack2019). Poor construct validity is likely due in part to the fact that indices derived from a given executive function task are known to be influenced by multiple construct-irrelevant sources of variability, including measurement artifacts associated with participants’ level of response caution (e.g., speed/accuracy tradeoffs) and motor response speed (Hedge et al., Reference Hedge, Vivian-Griffiths, Powell, Bompas and Sumner2019; Lerche & Voss, Reference Lerche and Voss2019; Stafford et al., Reference Stafford, Pirrone, Croucher and Krystalli2020).
An emergent alternative to this “fractionation” perspective uses formal mathematical models to describe the neurocognitive mechanisms that underlie task performance. Evidence accumulation models are a well-validated framework from mathematical psychology and computational neuroscience that explain individuals’ choices and RTs on a wide array of behavioral tasks (including executive function tasks). The models posit a process in which people gather evidence for possible responses until a critical evidence threshold is reached for a given choice (Smith & Ratcliff, Reference Smith and Ratcliff2004; Heathcote & Matzke, Reference Heathcote and Matzke2022). The diffusion decision model (DDM; Ratcliff, Reference Ratcliff1978; Ratcliff et al., Reference Ratcliff, Smith, Brown and McKoon2016), one of the most commonly used evidence accumulation models, decomposes task performance into parameters representing several mechanisms of influence on behavioral performance: (1) efficiency of evidence accumulation toward the correct choice (“drift rate”; v), (2) degree of response caution (boundary separation; a), (3) response biases, which tend to favor decisions with higher probabilities (start point; z), and (4) processes unrelated to the decision process, such as time for stimulus encoding and motor response execution (nondecision time; Ter; Voss et al. Reference Voss, Nagler and Lerche2013). The DDM has previously been used to measure these mechanisms on the SST by accounting for construct-irrelevant influences on SST performance (Epstein et al., Reference Epstein, Karalunas, Tamm, Dudley, Lynch, Altaye and Atluri2023; Fosco et al., Reference Fosco, Kofler, Alderson, Tarle, Raiker and Sarver2019b; Karalunas & Huang-Pollock, Reference Karalunas and Huang-Pollock2013).
Application of the DDM to large task batteries has repeatedly found that the model’s drift rate parameter forms a cohesive general factor that drives individual differences in performance across many complex cognitive tasks and shows strong psychometric properties, including good test-retest reliability (Eisenberg et al., Reference Eisenberg, Bissett, Zeynep Enkavi, Li, MacKinnon, Marsch and Poldrack2019; Lerche et al., Reference Lerche, von Krause, Voss, Frischkorn, Schubert and Hagemann2020; Schubert et al., Reference Schubert, Frischkorn, Hagemann and Voss2016; Schmiedek et al., Reference Schmiedek, Oberauer, Wilhelm, Süß and Wittmann2007). In contrast to assumptions of the “fractionation” perspective, recent work demonstrates that the vast majority of the variance in individuals’ performance on a variety of tasks thought to measure specific executive functions can instead be explained by this task-general factor (Löffler et al., Reference Löffler, Frischkorn, Hagemann, Sadus and Schubert2024; Weigard et al., Reference Weigard, Clark and Sripada2021; Weigard & Sripada, Reference Weigard and Sripada2021). As drift rate can be conceptualized as individuals’ efficiency of selectively gathering goal-relevant evidence to make adaptive choices across contexts, it also has strong conceptual links to traditional definitions of executive function (Weigard & Sripada, Reference Weigard and Sripada2021). Taken together there is strong evidence to support the drift rate parameter as an index of task-general executive functioning. Accordingly, the current study applied the DDM to the SST to characterize general executive functioning (Weigard et al., Reference Weigard, Soules, Ferris, Zucker, Sripada and Heitzeg2020; Zucker et al., 1996, 2000).
Reciprocal relationships
There is considerable empirical evidence that relationships between executive functioning, externalizing symptoms, and alcohol and drug use operate bidirectionally. Indeed, past work has reported that healthy levels of executive functioning are protective against a wide range of adverse psychosocial outcomes (Duckworth, Reference Duckworth2011; Royall et al., Reference Royall, Lauterbach, Cummings, Reeve, Rummans, Kaufer, LaFrance and Coffey2002), including externalizing symptoms (Alessandri et al., Reference Alessandri, Perinelli, Filosa, Eisenberg and Valiente2022; Hentges et al., Reference Hentges, Weaver Krug, Shaw, Wilson, Dishion and Lemery-Chalfant2020; Wiker et al., Reference Wiker, Pedersen, Ferschmann, Beck, Norbom, Dahl, von Soest, Agartz, Andreassen, Moberget, Westlye, Huster and Tamnes2023), and alcohol and drug use (Nigg et al., Reference Nigg, Wong, Martel, Jester, Puttler, Glass, Adams, Fitzgerald and Zucker2006; Piehler et al., Reference Piehler, Véronneau and Dishion2012). There is also evidence of bidirectional relationships between social adaptation and problem behavior, with healthy social adjustment predicting lower levels of externalizing symptoms and substance use across adolescence (Burke et al., Reference Burke, Pardini and Loeber2008; Jacobson & Crockett, Reference Jacobson and Crockett2000; Prinstein & La Greca, Reference Prinstein and La Greca2004). Although not central to the current study, we examine possible reciprocal relationships between task-general executive functioning, externalizing symptoms, alcohol and drug use, and social adaptation across parents, peers, and school.
The current study
The current study uses data from the Michigan Longitudinal Study (MLS). Of note, the sample is enriched for children of parents with an Alcohol Use Disorder (AUD), who are at risk to exhibit high levels of early adolescent alcohol use and externalizing psychopathology and low levels of executive functioning (Meldrum et al., Reference Meldrum, Trucco, Cope, Zucker and Heitzeg2018; Puttler et al., Reference Puttler, Fitzgerald, Heitzeg and Zucker2017). Greater representation of the high-end of problem behavior will strengthen our ability to detect the proposed associations (Paige et al., Reference Paige, Meisel and Colder2021). The hypotheses, design, and analyses of the current study were preregistered: https://osf.io/75zwg/?view_only = 46d2a2b0aa8d485d8acb860f4bdc1044. Some deviations from the preregistration should be noted. Specifically, we initially proposed to utilize several measures of executive functioning, including two behavioral tasks (the go/no-go in addition to the SST) as well as neuroimaging measures, in order to explore how hypothesized associations may differ across measures of executive functioning. After accessing the full dataset from the MLS, the sample sizes of the SST data (N range = 360–553) were substantially larger than the go/no-go (N range = 106–532) and neuroimaging (N range = 93–214) data across the four waves. Additionally, our own psychometric work with the neuroimaging data raised concerns about its reliability and validity. We elected to publish this psychometric work (Paige et al., Reference Paige, Colder, Cope, Hardee, Heitzeg, Soules and Weigard2024) separately and focus here on the central aim of the preregistered study, testing the ways in which adolescent alcohol use, externalizing symptoms, and social adaptation impact the development of executive functioning.
Aim 1
Test longitudinal bifactor models of externalizing symptoms from adolescence to early adulthood, which provides a way to separate alcohol use specific variance from general externalizing symptoms at four assessments spanning early adolescence to early adulthood and to examine whether the structure of co-occurring and unique symptoms vary across age. We hypothesize that a latent factor characterized by general externalizing symptoms and domain specific factors for alcohol use, drug use, aggressive behaviors, and rule-breaking behaviors will emerge at each assessment (see Figure 1).
Figure 1. Conceptual model of bifactor model for externalizing symptoms (Aim 1). Note. To reduce the number of model parameters, the rule-breaking and aggressive behavior subscales of the Achenbach measures were each parceled into five bundles at each wave.
Aim 2
Test prospective associations between general externalizing symptoms and domain specific facets and task-general executive functioning. We hypothesize that higher-order factors of externalizing symptoms will be related to poor task-general executive functioning. We also hypothesize that the specific factors of alcohol use, drug use, aggression, and rule-breaking will be related to poor task-general executive functioning.
Aim 3
Test potential indirect associations that may underlie the relationship between externalizing symptoms and deficits in task-general executive functioning. The relationship between alcohol use (and perhaps drug use) and task-general executive functioning is thought to represent a direct association consistent with the neurotoxic effects of adolescent alcohol use on the developing brain (Squeglia et al., Reference Squeglia, Jacobus and Tapert2014). In contrast, associations between externalizing symptoms and poor task-general executive functioning are posited to operate through negative socio-environmental experiences (Fay-Stammbach et al., Reference Fay‐Stammbach, Hawes and Meredith2014). This aim tests potential indirect pathways involving parents, peers, and school (see Figure 2). We hypothesize that high levels of externalizing symptoms will be associated with disrupted parental monitoring, low peer connectedness, and poor school engagement, which in turn, will be related to poor task-general executive functioning.
Figure 2. Conceptual model depicting indirect pathways from general externalizing symptoms factor scores to task-general executive functioning operating through peer connectedness (Aim 3). Note. Peer connectedness is depicted here, and separate models were estimated for each proposed social adaptation variable (e.g., parental monitoring, peer connectedness, and school engagement).
Method
Participants
Risk level of the offspring in the MLS was varied through recruitment of a population-based sample that differed in level of Alcohol Use Disorder (AUD) among the fathers (Zucker et al., Reference Zucker, Fitzgerald, Refior, Puttler, Pallas, Ellis, Fitzgerald, Lester and Zuckerman2000). The highest risk group (41%) were drunk drivers with at least 0.15% blood alcohol levels who also met other ascertainment criteria. A moderate risk group (29%) were community members with an AUD, uncovered during the canvass for controls. All moderate and high-risk individuals had an AUD diagnosis, were coupled with a partner and had a son who was 3 to 5 years old. The low risk group (30%) was an ecologically comparable set of families with no parental AUD at the time of recruitment, who were recruited from neighborhoods where the high-risk participants lived. Families were excluded from the MLS if the target child displayed signs of fetal alcohol syndrome or the mother reported drinking during pregnancy. Full biological siblings were also included in the MLS if they were within eight years of age of the male target child. Due to initial MLS recruitment, the majority of the sample was male (61.7%) and White (80.7%). Among adolescents of color, the largest groups were African American (7.4%), Hispanic (5.1%), and Biracial (2.7%). In subsequent years, families were added to the MLS to increase demographic diversity. The current study uses a subset of the MLS sample selected to represent four critical developmental windows: early adolescence (ages 10–13), middle adolescence (ages 14–17), late adolescence (ages 18–21), and early adulthood (ages 22–25). Sample sizes for all data across the four waves are presented in Table 1. For simplicity, these four developmental windows will be described as Waves (W) 1, 2, 3, and 4 henceforth.
Table 1. Sample sizes
Note. We selected four waves to represent developmental periods of adolescence, each that spans multiple years. Therefore, some participants contributed more than one observation within each developmental period. In these cases, we selected the observation that was closest to the average age within that wave.
The larger psychosocial sample used in the current study was comprised of majority males (67.2%) and was majority non-Hispanic White (80.5%). Additionally, on average, parents of target adolescents had completed some college (48.8%), with an additional 24.3% of parents completing college and 13.5% of parents possessing postgraduate education.
All procedures were approved by an Institutional Review Board. If participants were under the age of 18, consent from caregivers and assent from adolescents were obtained prior to participation. Adolescents provided informed consent after age 18.
Procedure
The MLS involves repeated measures of alcohol use, drug use, behavioral and psychological functioning, environment, and psychopathology that were all assessed across the four developmental windows described above. Annual assessments occurred between ages 10 and 25 for a subset of variables, including all alcohol use, drug use, and psychosocial variables used in the current study. Participants also completed behavioral tasks, including the SST, in a lab setting.
Measures
Demographics
Participant sex (0 = male, 1 = female), racial/ethnic status (0 = people of color, 1 = Caucasian/non-Hispanic White), and parental education (1 = Less than High School, 2 = High School, 3 = Some College, 4 = Bachelors, 5 = Post. Grad.) were included as covariates. Socioeconomic resources is a multidimensional construct that includes parental education, household income, and neighborhood disadvantage (Ensminger & Fothergill, Reference Ensminger, Fothergill, Bornstein and Bradley2003). A recent well-powered, multisite study using longitudinal Adolescent Brain Cognitive Development Study (ABCD) data found that parental education (controlling for household income-to-needs and neighborhood disadvantage) exhibited strong, consistent effects on brain development (Sripada et al., Reference Sripada, Gard, Angstadt, Taxali, Greathouse, McCurry, Hyde, Weigard, Walczyk and Heitzeg2022). Therefore, the current study used highest reported parental education as a measure of socioeconomic resources.
Alcohol and drug use
The Drinking and Drug History Forms for Children/Youth and Adults were administered to provide developmentally appropriate assessments of alcohol, cannabis, cigarette, and illicit drug (e.g., cocaine, hallucinogens, and nonprescription opioids) use annually (Zucker et al., Reference Zucker, Fitzgerald and Noll1990, Reference Zucker, Fitzgerald and Noll1993). These measures assessed quantity, frequency, and variability of alcohol and drug use as well as past year alcohol and drug use problems. At ages 18–23, 64% of MLS participants met criteria for AUD, validating the high-risk nature of the sample (Puttler et al., Reference Puttler, Fitzgerald, Heitzeg and Zucker2017).
Externalizing symptoms
The Youth Self-Report (YSR) and Adult Self-Report (ASR) from the Achenbach System of Empirically Based Assessments were used to provide developmentally appropriate assessments of externalizing symptoms annually (Achenbach et al., Reference Achenbach, Dumenci and Rescorla2003; Achenbach & Rescorla, Reference Achenbach and Rescorla2001). Externalizing problems on the YSR and ASR included two subscales, rule-breaking and aggressive behaviors. Items that measured alcohol and drug use were removed from the externalizing dimension in order to avoid confounding with variables measured by the Drinking and Drug History Forms for Children/Youth and Adults. Internal consistencies of the externalizing scale on the YSR and ASR were good across W1-4 (α range = 0.87–0.88), and were fair to good across W1-4 for the rule-breaking (α range = 0.70–0.77) and aggressive behaviors (α range = 0.83–0.84) subscales.
Parenting context
The Parent Monitoring-Youth Form is a 10-item measure that was administered to children annually beginning at age 11 to measure the supervision and monitoring provided by parents or caretakers outside of the school environment (Chilcoat, & Anthony, Reference Chilcoat and Anthony1996). It was adapted for age appropriateness by the MLS project from work by Patterson and Capaldi (Reference Patterson and Capaldi1988). In order to be developmentally appropriate, this measure was administered until targets were 18-years-old. Therefore, the current study only had parental monitoring data across W1-3. Parental monitoring was assessed through seven items that asked the target child to respond how often their parents are involved, are present with them, or how often their parents are involved in monitoring their behavior or whereabouts. Sample items included, “How often, before you go out, do you tell your [mom/dad/mom and dad/caregiver] when you will be back?” and “If you are at home when your parents are not, how often do you know how to get in touch with them?”. Responses were rated on a 5 point Likert scale (1 = All of the time, 5 = Never), and were reverse-coded in the current study so that higher scores indicated higher parental monitoring. Adolescents could also respond 9 = if the item did not apply to them. These responses were coded as missing data. The seven items were averaged within wave to compute scale scores of parental monitoring. Internal consistencies were fair to good across W1-3 (α range = 0.68–0.88).
Peer context
The 10-item Peer Relationships subscale of the Self-Image Questionnaire for Young Adolescents was used to measure peer connectedness (Petersen et al., Reference Petersen, Schulenberg, Abramowitz, Offer and Jarcho1984). This measure was administered annually beginning at age 11 and continued through early adulthood, spanning W1-4 in the current study. Adolescents were asked to answer questions, “about the way you interact with your friends.” Sample items included, “I find it extremely hard to make friends,” and “I think that other people just do not like me.” Responses to the scale items range from 1 = “very strongly agree” to 6 = “very strongly disagree.” Responses were reverse-scored and scale scores were computed so that higher scores indicate higher levels of peer connectedness. Instrument validity has been demonstrated (Petersen et al., Reference Petersen, Schulenberg, Abramowitz, Offer and Jarcho1984), and internal consistencies were good across W1-4 (α range = 0.81–0.89).
School context
The Teacher Report Form (TRF; Achenbach & Rescorla, Reference Achenbach and Rescorla2001) was administered annually and was used to measure school engagement. The TRF was completed by the one of the participant’s primary teachers. This measure was administered until targets were 18-years-old; therefore, the current study measured school engagement across W1-W3. Throughout the MLS, the return rate for teacher participation was just under 90%. School engagement was assessed using four items that asked teachers to compare the target adolescent’s functioning to typical pupils of the same age. Sample items included, “how hard is he/she working?” and, “how happy is he/she?”. Responses to the scale items range from 1 = “much less” to 6 = “much more.” Internal consistencies were good across W1-3 (α range = 0.88–0.90).
Executive functioning
Stop-signal task
The Stop-Signal Task (SST; Logan, Reference Logan, Dagenbach and Carr1994), a computerized choice reaction time task designed to measure inhibition, was administered annually in the MLS. Participants were presented with an “X” or “O” on a computer screen and instructed to respond by pressing one of two keys as quickly as possible upon seeing the stimulus. They were also instructed to withhold response (stop themselves) upon hearing a tone (stop signal) following the stimulus on the screen. The stop signal occurred on 25% of the trials. A stochastic process occurred in which participants’ response on prior trials was used by the program to set stop signal delay (warning time) on subsequent trials, in order to maintain a successful inhibition rate of 50%. Four blocks (256 trials), following two practice blocks, were administered in the MLS. The first practice block of 32 trials was used to practice the “go” part of the task and to set participants expectancy to respond as rapidly as possible to the “X” or “O” choice task. The second practice block was used to teach them to recognize the auditory stop signal.
Initially, the DOS computer program was used to administer the SST. As the study continued, an EPrime version of the SST was used. About 21% of the current sample complete the SST using the earlier DOS computer program. We considered only including cases that completed the SST using EPrime; however, we were hesitant to reduce the sample size as well as exclude participants who were more likely to be female and people of color. Therefore, we elected to utilize all SST data and covary computer program, in addition to sex, racial/ethnic status, and parental education in all analyses using SST data.
Data analytic strategy
Diffusion decision model
Stop-signal task
The diffusion model was fit to SST data using functions from the Dynamic Models of Choice suite of R functions (Heathcote et al., Reference Heathcote, Lin, Reynolds, Strickland, Gretton and Matzke2019). The central four drift diffusion model parameters were estimated for each individual: drift rate (v SST ), response conservativeness (a SST ), nondecision time (Ter SST ), and response bias (z SST ). Between-trial variability in Ter was also estimated but ther other between-trial variability parameters were fixed to 0 given the difficulty of recovering these parameters at low trial numbers and evidence that they have a minimal impact on estimates of the main parameters (Lerche et al., Reference Lerche, Voss and Nagler2017). Individual-level Bayesian estimation was implemented using the differential evolution Markov chan Monte Carlo (DE-MCMC: Turner et al., Reference Turner, Sederberg, Brown and Steyvers2013) approach with broad, uninformative priors. Upon examination of the response bias (z SST ) variables, data at W1 and W3 were kurtotic. Two cases at W1 and one case at W3 fell more than 3 standard deviations below the mean, suggesting noncompliance with the task. These cases were removed, and the data were normally distributed. Drift rate (v SST ) represents efficiency of gathering goal-relevant evidence to make an adaptive choice; larger drift rate values indicate faster information accumulation or better task-general executive functioning. The a SST , Ter SST , and z SST parameters were included as covariates across all models predicting v SST .
Hypothesis testing
Analyses were performed in SAS ® 9.4 and Mplus Software (Muthén & Muthén, Reference Muthén and Muthén1998-2018; SAS Institute Inc., 2013). Sex, racial/ethnic status, and parental education were included as statistical control variables in regression models (Karreman et al, Reference Karreman, van Tuijl, van Aken and Deković2009; Squeglia et al., Reference Squeglia, Spadoni, Infante, Myers and Tapert2009; Windle, Reference Windle1990).
Aim 1
To test aim 1, confirmatory factor analysis with Robust Maximum Likelihood estimation was used to estimate bifactor models of externalizing symptoms at four assessments spanning early adolescence to early adulthood. Bifactor modeling allows for all items to load onto a general factor and domain specific factors, thereby separating cross-sectional alcohol use specific variance from general externalizing symptoms variance (Figure 1). The general factor was indicated by items on the YSR and ASR, and items on the Drinking and Drug History Forms. The alcohol use specific factor was indicated by five items which assess alcohol use in the past year: past year frequency of alcohol use, number of drinking days in the past six months, quantity of alcohol use on drinking days in the past six months, number of days past year binge-drinking, and a count variable of past year alcohol use problems. The drug use specific factor was also indicated by five items: past year frequency of cigarette use, past month quantity of cigarette use, past year frequency of cannabis use, a count variable of past year drug use problems, and a count variable of past year other illicit drug use (e.g., hallucinogens, cocaine, nonprescription opioids, etc.). To reduce the number of model parameters, the rule-breaking and aggressive behavior subscales of the Achenbach measures were each parceled into five bundles at each wave (Little et al., Reference Little, Rhemtulla, Gibson and Schoemann2013). Fit of the bifactor models was assessed using conventional (e.g., comparative fit index [CFI], root mean square error approximation [RMSEA], standardized root mean square residual [SRMR]) as well as alternative fit statistics developed for bifactor models (e.g., omega, omega hierarchical subscale (ωHS), and construct replicability (H); Dueber, Reference Dueber2017; Rodriguez et al., Reference Rodriguez, Reise and Haviland2016a, Reference Rodriguez, Reise and Haviland2016b). ω and ωs are the latent variable analogs to coefficient alpha for the general externalizing factor and specific factors, respectively. ωH differs from ω in that it only represents the variance from a single latent variable whereas ω is a function of all common variance. ωHS is an index of the reliability of the specific factors after partitioning out the variance attributable to the general factor. The only computational difference between ωS and ωHS is that in the numerator, the term associated with variance on the general factor is removed, leaving only the variance associated with the specific factor. ω, ωS, ωH, and ωHS values greater than .70 indicate acceptable reliability and values greater than .80 indicate good reliability (Cortina, Reference Cortina1993; Rodriguez et al., Reference Rodriguez, Reise and Haviland2016b; Santos, Reference Santos1999; Tavakol & Dennick, Reference Tavakol and Dennick2011). To evaluate fit of the bifactor model, values lower than .80 for Percentage of Uncontaminated Variance and greater than .70 for Explained Common Variance indicate that data are unidimensional enough to introduce bias in fitting a multidimensional model (Quinn, Reference Quinn2014; Reise et al., Reference Reise, Bonifay and Haviland2013). Values of construct replicability, H, > .70 indicate that measured items adequately represent the latent factors (Hancock & Mueller, Reference Hancock and Mueller2001; Rodriguez et al., Reference Rodriguez, Reise and Haviland2016a). Factor score determinacies > .70 indicate adequate correlation of factor scores with latent factors (Rodriguez et al., Reference Rodriguez, Reise and Haviland2016b). Nested tests were used to evaluate the invariance of the factor structure across time. In order to reduce model complexity, which is a common concern for longitudinal bifactor models (McElroy et al., Reference McElroy, Belsky, Carragher, Fearon and Patalay2018), factor scores were computed to be used in our cross-lagged MLM regression models.
Aim 2 (Figure 3)
To test aim 2, MLMs were estimated in PROC MIXED procedure in SAS 9.4 with maximum likelihood estimation (ML). MLM allows for varying times of assessments between participants and easily accommodates missing data (Tabachnick, & Fidell, Reference Tabachnick and Fidell2013). Repeated measures (level 1) were nested within participants (level 2), and data were arranged to test cross-lagged associations (e.g., externalizing symptoms, alcohol use, and aggression at W1 predicts drift rate at W2, controlling for W1 drift rate, and this structure was repeated through W4). The models included a random intercept, and sex, racial/ethnic status, and parental education as level 1 covariates. All independent variables were standardized at the sample level to aid in the interpretation of the results (Hamer & Simpson, Reference Hamer and Simpson2009). The MLMs that included drift rate as outcomes also included covariates from the DDM (nondecision time; Ter, response conservativeness; a, and response bias; z)Footnote 1 .
Figure 3. Conceptual model depicting multilevel model for testing main effects between general externalizing symptoms, specific alcohol use, specific aggression, social adaptation, and task-general executive functioning. Note. For simplicity, the random intercept and covariates are not depicted here. Sex, racial/ethnic status, parental education, computer version and remaining drift diffusion model parameters, nondecision time, response conservativeness, and response bias were included as covariates.
In order to examine bidirectional associations, the models were restructured to test task-general executive functioning (drift rate) as a prospective predictor of each symptom factor. Separate models were run for each symptom factor (general externalizing, alcohol use, drug use, and aggression).
Aim 3
MLM using the PROC MIXED procedure in SAS 9.4 with ML and Monte Carlo simulation with 20,000 samples was used to test aim 3 (Preacher & Selig, Reference Preacher and Selig2012). Cross-lagged models were arranged such that externalizing symptoms at W1 predicted peer connectedness at W2, controlling for W1 peer connectedness, and this structure was repeated through W4. Given that no data were collected on parental monitoring and school engagement at W4, as was developmentally appropriate, cross-lagged models to test indirect effects through parental monitoring and school engagement were arranged such that externalizing symptoms at W1 predicted peer connectedness at W2, controlling for W1 peer connectedness, and this structure was only repeated through W3. In order to examine univariate effects, separate models were estimated for each proposed social adaptation variable (e.g., parental monitoring, peer connectedness, and school engagement; MacKinnon, Reference MacKinnon, Rose, Chassin, Presson and Sherman2000). A model that included all the proposed indirect assocaitions was also estimated to examine unique effects. Then, a cross-lagged model was estimated such that parental monitoring, peer connectedness, and school engagement at W1 predicted drift rate at W2, controlling for W1 drift rate, and this structure was repeated through W4. All models included a random intercept and sex, racial/ethnic status, and parental education as level 1 covariates. All predictor variables were standardized (Tabachnick, & Fidell, Reference Tabachnick and Fidell2013). Estimates (e.g., A and B paths) were input into a Monte Carlo simulation to test indirect effects (Preacher & Selig, Reference Preacher and Selig2012).
Results
Descriptive statistics
At W1, 96% of adolescents endorsed no alcohol use, which is not surprising given the age of the sample (10–13-years-old). Drinking increased across adolescence and early adulthood, with 30.3%, 71.6% and 86.3% of participants reporting past year alcohol use across Waves 2, 3, and 4, respectively. Specifically, at W2 (ages 14 to 17) about 13.7% of adolescents reported less than monthly drinking, 4.5% drank monthly, 11.3% drank about once a week, and less than 1% drank more than once a week. At W3 and W4, adolescents and early adults reported drinking 4.07 and 6.62 days per month on average, respectively. Alcohol use problems also increased across development. Whereas only 1.2% of adolescents at W1 endorsed any problems, this number increased to 21.7%, 48%, and 56.5% across waves 2, 3, and 4, respectively. Regarding illicit drug use, as expected, use was low at W1; 98.8% and 96.7% of adolescents reported no cannabis and no cigarette use, respectively. Cannabis use increased across adolescence, with 22.0% (W2) and 43.7% (W3) of adolescents reporting past year cannabis use, and then decreased slightly in early adulthood (38.5% endorsed past year use at W4). Endorsement of past year cigarette use increased across waves 2 (19.3%), 3 (41.9%), and 4 (46.4%). Experiencing problems related to illicit drug use was uncommon; 98.4%, 91.9%, 92.7%, and 96.8% endorsed 0 drug problems across waves 1, 2, 3, and 4, respectively.
Bifactor models
At W1 (ages 10–13), endorsement of illicit drug use was very low; 4.26% and 1.26% of the sample endorsed any cigarette and cannabis use in the past year, respectively. Therefore, these data were highly skewed and kurtotic, and we did not include drug use items in bifactor modeling at W1. Additionally, only 0.17% of the sample endorsed past year binge-drinking at W1, and this item was omitted from bifactor modeling at W1. Bifactor models of externalizing symptoms were estimated at each wave. Initial estimation of bifactor models produced a negative residual variance for the rule-breaking specific factors. A negative residual variance in this context suggests that after accounting for the general externalizing symptoms factor there was little variability left in the rule-breaking specific factors (Bollen, Reference Bollen1989). Removing the rule-breaking specific factors and allowing the rule-breaking items to load only on the general externalizing factors led the models to estimate without negative variance estimates. Rule-breaking specific factors likewise produced negative residual variances in our previous work utilizing a community sample of adolescents to estimate bifactor models of externalizing symptoms (Paige et al., Reference Paige, Meisel and Colder2021). Results for individual bifactor models estimated at each wave (W1-W4) are presented in Table 2.
Table 2. Bifactor results for single wave and multiple wave models
Note. CFI = comparative fit index. TLI = Tucker-Lewis index. RMSEA = root mean square error.
Approximation. SRMR = standardized root mean square residual. W = wave.
Then, the independent measurement models at each wave were combined into an overall measurement model with factors correlated across time. Details regarding longitudinal measurement invariance of bifactor models can be found in Supplemental Material 1. Fit information of final models is presented in Table 2. The final longitudinal externalizing symptoms bifactor measurement model including W1-W4 provided acceptable fit to the data (χ 2 = 4059.59(2500), p < .001, CFI = .94, TLI = .94, RMSEA = .02, SRMR = .07). Standardized factor loadings and alternative fit statistics across all bifactor models are presented in Table 3.
Table 3. Standardized factor loadings for bifactor models across waves 1, 2, 3, and 4
Note. G = General externalizing factor, W = Wave, AU = Alcohol Use, Freq=Frequency, AB = Aggressive Behavior, RB = Rule-breaking. ECV = Explained common variance. FR = Factor Determinacy.
The alcohol use item loadings ranged from 0.21 to 0.51 and from 0.55 to 0.93 on the general externalizing factors and alcohol use specific factors, respectively, across W1-4. The drug use item loadings had similar weak loadings on the general externalizing factor across W2-4 (range = 0.24–0.53), and they also had weak loadings on the drug use specific factors (range = 0.26–0.55), with the exception of the annual cannabis use frequency item (range = 0.61–0.74). Regarding aggression, the item loadings generally fell within the acceptable range on the general externalizing factors (range = 0.43–0.67). Consistent with our previous work estimating bifactor models of externalizing symptoms (Paige et al., Reference Paige, Meisel and Colder2021), the aggression item loadings on the specific factors were weak across early and middle adolescence, and were stronger across late adolescence and early adulthood (see Table 3). Finally, with the exception of one weak loading in early adolescence, rule-breaking behavior item loadings on the general externalizing symptoms factor were acceptable, ranging from 0.52 to 0.73.
In sum, the conventional and alternative fit statistics suggest that the longitudinal bifactor model provided an adequate fit to the data. The general externalizing factors and alcohol use specific factors were well specified (strong factor loadings), while the drug use and aggression specific factors were not (weak factor loadings), suggesting that there was less common variance in these specific factors after accounting for variance in general externalizing symptoms. Additional fit indices for the bifactor models are provided in Supplemental Material 1.
Main effects of externalizing symptoms, specific alcohol and drug use, and aggression predicting task-general executive functioning
Main effects MLMs were run to test prospective main effects of externalizing symptoms, specific alcohol and drug use, and aggression on task-general executive functioning across adolescence and emerging adulthood, controlling for prior task-general executive functioning and demographic variables (e.g., sex, racial/ethnic status, and parental education; Path C; see Figure 2). The results of main effects models for externalizing symptoms, specific alcohol useFootnote 2 and aggression predicting task-general executive functioning are reported in Table 4 (see Model 1).
Table 4. Multilevel modeling results for general externalizing symptoms, specific alcohol use, specific aggression, and social adaptation predicting task-general executive functioning (v SST )
Note: Model 1 indicates the main effects model for general externalizing symptoms, specific alcohol use, and specific aggression predicting task-general executive functioning (v SST ). Models 2-5 display the step-wise process of adding the social adaptation variables in order to examine univariate effects. Standardized coefficients are presented, with the exception of the drift diffusion covariates, which were included concurrently with the dependent variables. Racial/ethnic status was coded such that 0=People of color and 1= Caucasian/non−Hispanic White. EF = Executive functioning. Bolded values = p < .001, ** = p < .01, * = p < .05.
Results from the intercept only model revealed that the ICC was 0.5185, indicating that 51.85% of the variance for drift rate during SST was due to clustering within individuals. This represents fair test-retest reliability by conventional standards, but relatively robust given the average measurement interval of 2–3 years (whereas test-retest reliability is typically estimated across much shorter intervals)Footnote 3 . Stability of task-general executive functioning (v SST ) across adolescence and emerging adulthood was supported (β = 0.43, p < .001; see Table 4). Regarding covariates, parental education (β = 0.09, p < .001) was related to task-general executive functioning (v SST ), such that adolescents whose parents reported high levels of education had higher levels of task-general executive functioning (v SST ). Nondecision time (Ter SST ) was positively associated with task-general executive functioning (v SST ), while response conservativeness (a SST ) was negatively related to task-general executive functioning (v SST ). General externalizing symptoms, specific alcohol use, and specific aggression did not significantly predict task-general executive functioning (v SST ) across timeFootnote 4 .
Main effects of social adaptation predicting task-general executive functioning
Next, we added social adaptation (i.e., parental monitoring, peer connectedness, and school engagement) to the main effects MLM to test prospective main effects on task-general executive functioning across adolescence and emerging adulthood, controlling for prior task-general executive functioning and demographic variables (e.g., sex, racial/ethnic status, and parental education; Path B; see Figure 2). These variables (i.e., parental monitoring, peer connectedness, and school engagement) were first tested in separate models and then tested simultaneously in a combined model. This allowed examination of unadjusted and adjusted associations. The results of main effects models for parental monitoring, peer connectedness, and school engagement predicting task-general executive functioning are reported in Table 4 (see Models 2-5).
When estimated in separate models, peer connectedness (β = 0.07, p < .01), and school engagement (β = 0.14, p < .001) prospectively predicted task-general executive functioning (v SST ), such that high levels of peer connectedness and school engagement were related to better task-general executive functioning (v SST ) across adolescence and emerging adulthood, controlling for previous levels of task-general executive functioning (v SST ) and covariates. The prospective association between parental monitoring and task-general executive functioning (v SST ) was nonsignificant.Footnote 5
Next, an MLM with all social adaptation variables predicting task-general executive functioning (v SST ) was estimated to examine univariate effects (see Table 4; Model 5). In this model, peer connectedness (β = 0.10, p < .01) and school engagement (β = 0.14, p < .001) still significantly predicted task-general executive functioning (v SST ), suggesting that these social adaptation variables had unique effects on task-general executive functioning.
Main effects of general externalizing symptoms, specific alcohol use, and aggression predicting social adaptation
Main effects MLMs were run to test prospective main effects of general externalizing symptoms, specific alcohol use, and aggression predicting social adaptation (i.e., parental monitoring, peer connectedness, and school engagement) across adolescence and emerging adulthood, controlling for prior social adaptation, and covariates (e.g., sex, racial/ethnic status, and parental education; Path A; see Figure 2). The results of main effects models for general externalizing symptoms, specific alcohol use, and aggression predicting social adaptation are reported in Table 5.
Table 5. Multilevel modeling results for general externalizing symptoms, specific alcohol use, and specific aggression predicting social adaptation
Note: Sex was coded such that 0=male, 1=female. Racial/ethnic status was coded such that 0=People of color and 1= Caucasian/non-Hispanic White. Standardized coefficients are presented. Bolded values = p < .001, ** = p < .01, * = p < .05.
Parental monitoring
Stability of parental monitoring across adolescence was supported (β = 0.24, p < .001). Sex, racial/ethnic status, and parental education were not related to parental monitoring. Regarding hypothesized pathways, general externalizing symptoms prospectively predicted low levels of parental monitoring across adolescence (β = −0.14, p < .001). Specific alcohol use and specific aggression did not predict changes in parental monitoring.
Peer connectedness
Stability of peer connectedness across adolescence and early adulthood was supported (β = 0.35, p < .001). Regarding covariates, sex, racial/ethnic status, and parental education were not related to peer connectedness. With respect to hypothesized pathways, high levels of general externalizing symptoms (β = −0.05, p < .01) and specific aggression (β = −0.05, p < .01) prospectively predicted low levels of peer connectedness across adolescence and early adulthood. Additionally, domain specific alcohol use predicted changes in peer connectedness across time in the unexpected direction, such that high levels of specific alcohol use predicted high peer connectedness (β = 0.04, p < .05).
School engagement
Stability of school engagement across adolescence was supported (β = 0.58, p < .001). Regarding covariates, sex (β = 0.16, p < .01) and parental education (β = 0.26, p < .001) were significantly related to school engagement, such that girls and adolescents whose parents reported high levels of education had high levels of school engagement. With respect to hypothesized pathways, high levels of general externalizing symptoms prospectively predicted low levels of school engagement across adolescence (β = −0.21, p < .01). Specific alcohol use and specific aggression were not prospectively related to school engagement.
Indirect effects
A conceptual figure depicting significant indirect effects is provided in Figure 4.
Figure 4. Conceptual model depicting significant indirect effects between general externalizing symptoms, specific alcohol use, specific aggression, social adaptation, and task-general executive functioning. Note. Significant pathways are presented here. Blue paths indicate negative direction of effects, while green paths indicate positive direction of effects. Multilevel models predicting each social adaptation variables (e.g., parental monitoring, peer connectedness, and school engagement) were run separately, but significant results are depicted together here in a conceptual figure. Social adaptation only significantly predicted changes in task-general executive functioning as measured by drift rate on the stop signal task. vSST = drift rate on the stop signal task.
Parental monitoring
Given that parental monitoring did not prospectively predict task-general executive functioning, there were no plausible indirect effects involving parental monitoring.
Peer connectedness
Given the pattern of main effects reported above, a plausible indirect pathway was from general externalizing symptoms to task-general executive functioning (v SST ) through peer connectedness. There were significant, prospective indirect effects from externalizing symptoms to task-general executive functioning (v SST ) through peer connectedness (95% CI[−0.076, −0.007]). High levels of general externalizing symptoms predicted poor peer connectedness, which in turn, predicted low levels of task-general executive functioning (v SST ).
Further, the pattern of main effects suggested an indirect pathway from specific aggression to task-general executive functioning (v SST ) through peer connectedness. There were significant, prospective indirect effects from specific aggression to task-general executive functioning (v SST ) through peer connectedness (95% CI[−0.143, −0.010]). In other words, high levels of specific aggression predicted poor peer connectedness, which was subsequently related to low levels of task-general executive functioning (v SST ).
Additionally, results from multilevel models suggested an indirect pathway from specific alcohol use to task-general executive functioning (v SST ) through peer connectedness. There were significant, prospective indirect effects from specific alcohol use to task-general executive functioning (v SST ) through peer connectedness (95% CI[0.001, 0.031]). The direction of this indirect effect was somewhat surprising; high levels of specific alcohol use predicted better peer connectedness, which subsequently predicted high levels of task-general executive functioning (v SST ).
School engagement
Given the pattern of findings above, another plausible indirect pathway was from general externalizing symptoms to task-general executive functioning (v SST ) through school engagement. There were significant, prospective indirect effects from externalizing symptoms to task-general executive functioning (v SST ) through school engagement (95% CI[−0.277, −0.040]). Results suggest that high levels of externalizing symptoms predicted poor school engagement, which was subsequently related to low levels of task-general executive functioning (v SST ).
Reciprocal relationships
Main effects MLMs were run to test prospective main effects of task-general executive functioning and social adaption on general externalizing symptoms, specific alcohol use, and specific aggression across adolescence and emerging adulthood, controlling for prior levels of the dependent variable and demographic variables (e.g., sex, racial/ethnic status, and parental education). The results of models testing reciprocal relationships are reported in Table 6.
Table 6. Multilevel modeling results for task-general executive functioning and social adaptation predicting general externalizing symptoms, specific alcohol use, specific drug use, and specific aggression
Note: Sex was coded such that 0=male, 1=female. Racial/ethnic status was coded such that 0=People of color and 1= Caucasian/non-Hispanic White. Standardized coefficients are presented. Because results from bifactor modeling did not support a specific factor of drug use at wave 1, we estimated a separate multilevel model with data arranged to test cross-lagged associations predicting specific drug use (task-general executive functioning and social adaptation at wave 2 predicts specific drug use at wave 3, controlling for wave 2 specific drug use, and this structure was repeated through wave 4, allowing for the test of two cross-lags). Bolded values = p < .001, ** = p < .01, * = p < .05.
General externalizing symptoms
General externalizing symptoms were stable across adolescence and emerging adulthood (β = 0.14, p < .001). Racial/ethnic status (β = 0.01, p < .05) was related to general externalizing symptoms, such that non-Hispanic White adolescents had higher levels of externalizing symptoms, but this was a very small effect. Task-general executive functioning (v SST ) did not predict externalizing symptoms across time. Social adaption (parent monitoring, peer connectedness, and school engagement) was unrelated to externalizing symptoms.
Specific alcohol use
Specific alcohol use was stable across time (β = 0.13, p < .001). Racial/ethnic status (β = 0.07, p < .001) was related to specific alcohol use, such that non-Hispanic White adolescents endorsed higher levels of specific alcohol use. Task-general executive functioning (v SST ) did not prospectively predict specific alcohol use. With respect to social adaptation, peer connectedness prospectively predicted specific alcohol use across adolescence and emerging adulthood (β = 0.05, p < .01). Here again, this effect was in an unexpected direction, such that high peer connectedness was related to high levels of specific alcohol use.
Specific aggression
Specific aggression was stable across adolescence and emerging adulthood (β = 0.08, p < .001). Regarding covariates, sex (β = 0.01, p < .01) was related to specific aggression, such that females exhibited higher levels of specific aggression. Task-general executive functioning (v SST ) did not prospectively predict changes in specific aggression across time. High levels of school engagement predicted low levels of specific aggression (β = −0.009, p < .05).
Specific drug use
Because results from bifactor modeling did not support a specific factor of drug use at W1, we estimated a separate MLM with data arranged to test cross-lagged associations predicting specific drug use (v SST and social adaptation at W2 predicts specific drug use at W3, controlling for W2 specific drug use), and this structure was repeated through W4, allowing for testing of two cross-lags. Specific drug use was stable across adolescence and emerging adulthood (β = 0.05, p < .001). Regarding covariates, racial/ethnic status (β = 0.01, p < .001) was related to specific drug use, such that non-Hispanic White adolescents endorsed higher levels of specific drug use. Task-general executive functioning (v SST ) did not predict specific drug use. Social adaptation across parents, peers, and school did not prospectively predict specific drug use.
Discussion
The existing literature has yielded inconsistent findings regarding whether the effects of alcohol use during adolescence, a sensitive period of brain development, are neurotoxic and derail healthy cognitive development (Lees et al., Reference Lees, Mewton, Stapinski, Squeglia, Rae and Teesson2019). Notably, previous studies have failed to consider that adolescent alcohol use occurs within a broader developmental context of externalizing symptoms (including other drug use) and poor social development. Additionally, fractionated approaches and heterogeneity in the measurement of executive functioning (Nigg, Reference Nigg2017) may contribute to mixed findings. This study aimed to provide clarity to the alcohol use-executive functioning association by (1) distinguishing general externalizing symptoms from domain specific symptoms (alcohol use, aggression, rule-breaking, drug use) using bifactor modeling, (2) examining prospective associations between general externalizing symptoms and domain specific symptoms, and task-general executive functioning across adolescence and early adulthood, (3) using computational modeling of performance on the SST to provide a more psychometrically sound index of task-general executive functioning, and (4) testing indirect pathways from externalizing symptoms to deficits in task-general executive functioning through poor social adaptation across parents, peers, and school. A large high-risk longitudinal sample enriched for children of parents with an AUD (the Michigan Longitudinal Study; Zucker et al., Reference Zucker1996) was used to test study aims.
Aim 1: hierarchical structure of externalizing symptoms
In line with calls to examine developmental changes in the structure of psychopathology (Forbes et al., Reference Forbes, Tackett, Markon and Krueger2016), the current study estimated longitudinal bifactor models of externalizing symptoms across early adolescence (age 10) to early adulthood (age 25). Our first hypothesis that a latent factor characterized by general externalizing symptoms and domain specific factors for alcohol use, drug use, aggressive behaviors, and rule-breaking behaviors would emerge at each assessment was largely supported. Results from bifactor modeling supported a hierarchical structure of externalizing psychopathology at four assessments spanning early adolescence to early adulthood; a general externalizing factor and specific factors emerged at each assessment. This is consistent with a wide body of research on the Hierarchical Taxonomy of Psychopathology (HiTOP; Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby, Brown, Carpenter, Caspi, Clark, Eaton, Forbes, Forbush, Goldberg, Hasin, Hyman, Ivanova, Lynam, Markon and Zimmerman2017), and supports the notion that substance use is part of a larger constellation of problem behavior (Krueger et al., Reference Krueger, Hobbs, Conway, Dick, Dretsch, Eaton, Forbes, Forbush, Keyes, Latzman, Michelini, Patrick, Sellbom, Slade, South, Sunderland, Tackett, Waldman and Waszczuk2021).
Specific factors of alcohol use and aggression emerged at each wave. However, specific factors of drug use did not emerge until Wave 2 (ages 14–17), due to very low endorsement of drug use at Wave 1 when the sample was 10–13-years-old. This finding is novel as no other studies have considered such a bifactor model starting in early adolescence, and results support Forbes et al. (Reference Forbes, Tackett, Markon and Krueger2016) contention that exposure to environmental risks (e.g., drugs) guides the manifestation of increasingly specific factors of externalizing symptoms across development. It should also be noted that while the general externalizing and specific alcohol use factors represented the data well and were reliable, the specific factors of aggression and drug use were less well-represented and less reliable. This pattern replicates our prior work which found poorly specified specific factors of aggression and substance use compared to general factors of externalizing symptoms in an adolescent sample (Paige et al., Reference Paige, Meisel and Colder2021). Additionally, due to the poorly represented and less reliable specific factors, results concerning specific factors of aggression and drug use should be interpreted with caution.
Specific factors of rule-breaking behavior did not emerge. This is somewhat inconsistent with theory that behaviors like delinquency are distinguishable from general externalizing symptoms and other specific externalizing factors in adolescence (Forbes et al., Reference Forbes, Tackett, Markon and Krueger2016; Krueger & Tackett, Reference Krueger, Tackett, Beauchaine and Hinshaw2014). However, in a separate sample, we similarly found no evidence specific factors of rule-breaking behaviors in a longutidinal bifactor model (Paige et al., Reference Paige, Meisel and Colder2021). Additionally, factor analytic work by Krueger et al. (Reference Krueger, Markon, Patrick, Benning and Kramer2007) suggests that externalizing symptoms are characterized by a hierarchical structure that includes a general externalizing factor and two lower-order factors of aggression and substance use. Our results were consistent with this model, suggesting that the majority of rule-breaking variance is accounted for by factors of general externalizing symptoms, and there is likely little variance “leftover” to specify domain specific factors of rule-breaking behaviors (Krueger et al., Reference Krueger, Markon, Patrick, Benning and Kramer2007). In conclusion, findings from the current study support the notion that externalizing symptoms are best understood through a hierarchical structure, including a general factor characterized by rule-breaking, aggression, and substance use behaviors as well as domain specific factors of aggression, alcohol use, and drug use, which emerges as adolescents age.
Aim 2: direct effects of externalizing symptoms, specific alcohol and drug use, and aggression on the development of task-general executive functioning
Direct effects of general externalizing symptoms on task-general executive functioning
General externalizing symptoms were hypothesized to be related to poor task-general executive functioning as a direct effect and this was not supported. This is somewhat inconsistent with prior studies that have reported prospective associations between problem behavior and questionnaire-assessed executive functioning (Atherton et al., Reference Atherton, Zheng, Bleidorn and Robins2019; Paige et al., Reference Paige, Meisel and Colder2021; Wiker et al., Reference Wiker, Pedersen, Ferschmann, Beck, Norbom, Dahl, von Soest, Agartz, Andreassen, Moberget, Westlye, Huster and Tamnes2023). However, measurement of executive functioning in the current study differed from this prior work. While Fosco et al. (Reference Fosco, Hawk, Colder, Meisel and Lengua2019a) reported links between SSRT and subsequent delinquency across adolescence, no previous studies, to our knowledge, have examined the consequential effects of problem behavior on behavioral performance on the SST.
It should be noted that empirical evidence linking externalizing symptoms to executive functioning during adolescence and early adulthood is lacking. This is especially surprising, given the developmentally normative increase in deviant behavior (Moffitt, Reference Moffitt1993) as well as critical changes in neurodevelopment (Blakemore & Choudhury, Reference Blakemore and Choudhury2006) that occur during this developmental period. It may be that the small number of published studies in this area is due to nonsignificant associations between externalizing symptoms and executive functioning across adolescence (e.g., the “file drawer problem”; Rosenthal, Reference Rosenthal1979). While nonsignificant direct effects from the current study support this notion, importantly, findings suggest that the broader problem behavior context still confers risk for poor executive functioning (v SST ) development, and indirect pathways through social adaption are discussed in detail under Aim 3.
Neurotoxic effects of alcohol use on task-general executive functioning
Our third hypothesis that the alcohol use specific factors would be related to poor task-general executive functioning was not supported. Additionally, with respect to developmental timing, there was no evidence that alcohol use during the sensitive period of early adolescence (ages 10–13) had deleterious effects on the development of executive functioning. As noted earlier, the literature testing neurotoxicity effects of alcohol and executive function is limited and mixed. The current study accounted for the broader context of problem behavior and poor social development in which adolescent alcohol use often occurs, and also used a reliable measure of task-general executive functioning (drift rate), which may enhance construct validity in measurement of individual differences that underlie performance on executive functioning tasks (Löffler et al., Reference Löffler, Frischkorn, Hagemann, Sadus and Schubert2024; Weigard et al., Reference Weigard, Clark and Sripada2021; Weigard & Sripada, Reference Weigard and Sripada2021).
Indeed, as variance in specific alcohol use was separated from variance accounted for by general externalizing symptoms, specific drug use, and specific aggression, after accounting for these behaviors, remaining specific alcohol use variance may be an indicator of normative developmental trajectories. Alcohol use is normative in American culture (Johnston et al., Reference Johnston, Miech, O’Malley, Bachman, Schulenberg and Patrick2018), and some amount of use may be an indicator of healthy psychosocial adjustment in certain contexts (Peele & Brodsky, Reference Peele and Brodsky2000). Notably, non-Hispanic White adolescents had higher scores on the specific alcohol use factors. It is possible that these specific factors represented less harmful alcohol use that may be unique to non-Hispanic White Americans. Indeed, there is substantial empirical evidence that compared to non-Hispanic White Americans, Black Americans experience higher levels of alcohol problems despite later initiation of drinking and lower levels of use across almost all age groups, likely due to a complex interaction of factors related to discrimination, racism, and within-group social disapproval (Zapolski et al., Reference Zapolski, Pedersen, McCarthy and Smith2014). Further, there were positive associations between specific alcohol use and peer connectedness, discussed in more detail below. Notably, recent studies that have accounted for the broader context of problem behavior have reported that adolescent and emerging adult alcohol use is not correlated with executive functioning (Meehan et al., Reference Meehan, De Panfilis, Cain and Clarkin2013; Paige et al., Reference Paige, Meisel and Colder2021; Piehler et al., Reference Piehler, Véronneau and Dishion2012). It should be stated that evidence does not suggest that adolescent drinking directly promotes healthy adjustment, but rather, emphasizes the need to consider that adolescent alcohol use occurs in a broader context of problem behavior. In summary, it may be that the relationship between early alcohol use and poor executive functioning operates within a larger constellation of externalizing symptoms and socio-environmental experiences.
Neurotoxic effects of drug use on task-general executive functioning
As expected, there was lower variability in drug use in early adolescence compared to alcohol use (Swendsen et al., Reference Swendsen, Burstein, Case, Conway, Dierker, He and Merikangas2012), which presented some challenges in examining the unique effects of drug use on the development of task-general executive functioning. Indeed, the data did not support a specific factor of drug use in early adolescence. This limited examination of prospective effects of specific drug use on executive functioning to two cross-lags (middle to late adolescence, and late adolescence to early adulthood). Our hypothesis that the drug use specific factors would be related to poor task-general executive functioning was not supported. However, it should be noted that drug use specific factors were not well-defined by the indicators and had poor reliability, and so some caution is warranted in the conclusion that that early drug use does not have deleterious effects on long-term executive functioning. Notably, past work has largely failed to account for the broader context of problem behavior in which drug use occurs (Squeglia et al., Reference Squeglia, Spadoni, Infante, Myers and Tapert2009; Wetherill et al., Reference Wetherill, Squeglia, Yang and Tapert2013), and it is possible that previously observed direct links between drug use and poor executive functioning are actually due to co-occurring externalizing symptoms and poor social development.
Unique effects of aggression and rule-breaking behaviors on task-general executive functioning
There was no evidence that the specific factors of aggression and rule-breaking behaviors were related to low levels of task-general executive functioning. Bifactor modeling did not support estimation of specific factors of rule-breaking behavior, and this precluded examination of unique effects of rule-breaking behaviors on executive functioning. With respect to aggression, these specific factors were not well-defined by the indicators and had poor reliability. Further, results suggested that this domain specific factor likely did not represent physical aggression; as girls endorsed higher levels of specific aggression, and physical aggression has been reliably established to be more prevalent among males across development (Card et al., Reference Card, Stucky, Sawalani and Little2008). Some items on the YSR/ASR aggression subscale measure emotional lability (e.g. “My mood or feelings change suddenly”), while others measure forms of aggression that may not be as clearly violent (e.g., “I argue a lot”). After accounting for general externalizing variance, unique variance for the aggression specific factors may have been representative of emotional lability or hostility. Indeed, adolescent females are more likely to endorse emotional lability (Stringaris & Goodman, Reference Stringaris and Goodman2009). Additionally, results are consistent with our past work found no prospective associations between factors of specific aggression, which likely represented emotional lability or hostility, and effortful control across five assessments spanning ages 11–20 (Paige et al., Reference Paige, Meisel and Colder2021). No prior studies, to our knowledge, have examined the prospective effects of emotional lability on cognitive development as measured by behavioral tasks. This may be a useful direction for future research.
Aim 3: indirect pathways from general externalizing symptoms to task-general executive functioning through social adaptation
Indirect effects through parental monitoring
Our hypothesis that there would be indirect effects between externalizing symptoms and deficits in task-general executive functioning through parental monitoring was not supported. Indeed, parental monitoring was not related to task-general executive functioning (v SST ). This is inconsistent with recent longitudinal work by Hong et al. (Reference Hong, Deater-Deckard, Skinner, Duell, Lansford and Steinberg2024) which demonstrated that lower parental acceptance and higher rejection was associated with weaker improvements in adolescent executive functioning. Notably, no previous studies, to our knowledge, have examined the consequences of parental monitoring, specifically, on adolescent executive functioning. The current study suggests that that examining the impact of parenting practices related to parent-adolescent relationship quality, specifically, may be a more fruitful future direction.
Notably, general externalizing symptoms predicted changes in parental monitoring, such that high levels of externalizing symptoms predicted low levels of parental monitoring across adolescence. This is an important finding because research in this area has yielded mixed results, with some early childhood studies demonstrating that externalizing symptoms elicit parental overcontrol (Eisenberg et al., Reference Eisenberg, Taylor, Widaman and Spinrad2015). Conversely, cross-sectional associations in an adolescent sample suggest that externalizing symptoms elicit low parental monitoring, above and beyond prior levels of parental monitoring (Fite et al., Reference Fite, Colder, Lochman and Wells2006). This divergent pattern of associations between externalizing symptoms and parental monitoring emphasizes the importance of developmental timing when considering these trajectories. Indeed, Patterson’s (Reference Patterson, Reid, Patterson and Snyder2002) coercion model suggests that when faced with problem behavior, parents often initially respond by becoming increasingly involved in attempts to control their child’s behavior. However, over time, parents who are repeatedly faced with problem behavior and lack the skills to effectively intervene likely feel increasingly ineffective and become alienated from parenting (Patterson, Reference Patterson, Reid, Patterson and Snyder2002). By the time that children who exhibit high levels of problem behavior have reached adolescence, parents have likely reached this point in the coercion cycle, leading them to decrease their monitoring in response to elevated externalizing symptoms across time. Findings from the current study advance our understanding by demonstrating prospective effects between offspring externalizing symptoms and decreases in parental monitoring across early to late adolescence, supporting the notion that offspring problem behavior leads to parental disengagement across long periods of time.
Indirect effects through peer connectedness
Our hypothesis that there would be indirect effects between externalizing symptoms and deficits in task-general executive functioning through peer connectedness, such that high levels of externalizing symptoms would be related to low levels of peer connectedness, which in turn, would be related to low levels of task-general executive functioning was supported. It has been well-documented that externalizing symptoms elicit poor relationships with peers (Baumeister et al., Reference Baumeister, DeWall, Ciarocco and Twenge2005), and our findings are consistent with this literature. The significant, prospective association between poor peer connectedness and decreases in task-general executive functioning (v SST ) across adolescence and early adulthood is novel and important. Indeed, findings from the current study support the notion that adolescents who are rejected by prosocial peers may have fewer opportunities for the practice of social competence skills that involve executive functioning (Calkins, Reference Calkins1994; Fox & Calkins, Reference Fox and Calkins2003), and receive little feedback and guidance for their behavior in the peer context, which likely derails healthy cognitive development (v SST ; Farley & Kim-Spoon, Reference Farley and Kim-Spoon2014; Leflot et al., Reference Leflot, van Lier, Verschueren, Onghena and Colpin2011).
Additionally, while not specifically hypothesized, there was evidence to support indirect effects between specific aggression and poor task-general executive functioning (v SST ) through peer connectedness. The prospective effect between specific aggression, which likely represents emotional lability, and poor peer connectedness is consistent with the reliably documented effects of aggression-related emotional lability on social maladjustment (Kamper & Ostrov, Reference Kamper and Ostrov2013; Murray-Close, Reference Murray-Close2013; Ostrov, Reference Ostrov2008). The current study extends past work in this area, which has largely been concentrated in childhood, by demonstrating that emotional lability indirectly predicts task-general executive functioning through poor peer connectedness across four assessments that span ages 10 to 25. However, results bear replication due to poor reliability and low H values on the aggression specific factors. Future work in this area may aim to utilize scaled scores of emotional lability or hositility when investigating indirect effects from specific aggression to poor executive functioning through social maladaptation.
It should also be noted that specific alcohol use impacted the development of task-general executive functioning (v SST ) indirectly through peer connectedness; however, this indirect effect was in the opposite direction of what was expected. Indeed, high levels of specific alcohol use predicted better relationships with peers, which in turn, was subsequently related to high levels of task-general executive functioning (v SST ). Importantly, there is no evidence that adolescent alcohol use has positive, direct, neurobiological effects on cognitive development. Rather, our interpretation of the alcohol use specific factor is that when variance related to drinking is separated from the broader context of problem behavior, this specific variance is likely a marker of positive social adjustment in some contexts (Peele & Brodsky, Reference Peele and Brodsky2000). In other words, findings from the current study do not suggest that adolescent drinking is protective. Rather, results suggest that when using bifactor modeling to parse externalizing variance, variance related to drinking that is separate from variance related to rule-breaking, aggression, and other drug use is indicative of something positive, such as sociability.
Indirect effects through school engagement
Our hypothesis that there would be an indirect pathway between externalizing symptoms and deficits in task-general executive functioning through school engagement, such that high levels of externalizing symptoms would be related to poor school engagement, which in turn, would be related to low levels of task-general executive functioning was supported. Results corroborate substantial evidence that externalizing symptoms elicit poor school engagement throughout childhood (Stipek & Miles, Reference Stipek and Miles2008) and adolescence (Masten et al., Reference Masten, Roisman, Long, Burt, Obradović, Riley, Boelcke-Stennes and Tellegen2005). Moreover, our findings forward the idea that school engagement is critical for supporting cognitive development (Berry, Reference Berry2012), and provide novel empirical evidence that school engagement plays a central role in strengthening task-general executive functioning (v SST ) across ages 10–25. School provides a rich context for adolescents to observe modeled executive functioning skills, as well as practice behaviors related to healthy cognitive skills and receive direct and indirect feedback on these behaviors.
Reciprocal relationships
While not a primary aim of the current study, examination of reciprocal relationships was illuminating. Notably, drift rate did not predict externalizing symptoms across time, diverging from a large literature on the protective effects of healthy executive functioning in guarding against problem behavior (Alessandri et al., Reference Alessandri, Perinelli, Filosa, Eisenberg and Valiente2022; Fosco et al., Reference Fosco, Hawk, Colder, Meisel and Lengua2019a; Hentges et al., Reference Hentges, Weaver Krug, Shaw, Wilson, Dishion and Lemery-Chalfant2020; Wiker et al., Reference Wiker, Pedersen, Ferschmann, Beck, Norbom, Dahl, von Soest, Agartz, Andreassen, Moberget, Westlye, Huster and Tamnes2023). This study was the first, to our knowledge, to apply drift diffusion modeling to the SST across ages 10–25. Notably, the lack of significant effects in task-general executive functioning (v SST ) predicting general externalizing symptoms increases confidence in the notion that the direction of this within-person association flows from externalizing symptoms to poor task-general executive functioning (v SST ). This is consistent with a recent longitudinal study of adolescents that reported prospective associations between externalizing symptoms and poor executive functioning, but found no support for bidirectional associations (Brieant et al., Reference Brieant, King-Casas and Kim-Spoon2022). In contrast, analyses in the ABCD consistently find prospective relations between executive functioning and externalizing psychopathology, potentially due to the statistical power of this larger sample (Romer & Pizzagalli, Reference Romer and Pizzagalli2021; Wiker et al., Reference Wiker, Pedersen, Ferschmann, Beck, Norbom, Dahl, von Soest, Agartz, Andreassen, Moberget, Westlye, Huster and Tamnes2023). It is also relevant that, under ideal measurement conditions (e.g., multiple tasks), drift rate appears to be a relatively stable trait (Schubert et al., Reference Schubert, Frischkorn, Hagemann and Voss2016; Weigard et al., Reference Weigard, Clark and Sripada2021). If the rank ordering of individuals’ drift rate stays relatively stable across development, between-subjects associations of drift rate with psychopathology may reflect directional relations between these constructs at the level of individuals’ time-invariant traits, while within-subjects relations may be more subtle and difficult to detect. More work utilizing drift diffusion modeling to index task-general executive functioning when examining prospective associations between reaction time-based measures of executive functioning and problem behavior across adolescence is warranted.
Relatedly, drift rate did not significantly predict changes in specific factors of alcohol use, drug use, and aggression. This is somewhat inconsistent with empirical evidence that healthy cognitive development reduces levels of substance use and aggression across adolescence and early adulthood (Nigg et al., Reference Nigg, Wong, Martel, Jester, Puttler, Glass, Adams, Fitzgerald and Zucker2006; Piehler et al., Reference Piehler, Véronneau and Dishion2012). However, our interpretation of the alcohol use specific factor suggests that once variance unique to drinking is separated from problem behavior, it is likely an indicator of sociability. Indeed, previous studies that account for the broader context of problem behavior report nonsignificant associations between effortful control and low levels of adolescent alcohol use (Paige et al., Reference Paige, Meisel and Colder2021; Piehler et al., Reference Piehler, Véronneau and Dishion2012). Moreover, nonsignificant associations are consistent with our past work estimating bifactor models of externalizing symptoms and examining bidirectional associations with effortful control across adolescence, which suggests that healthy cognitive development does not prospectively predict specific factors of substance use nor aggression (Paige et al., Reference Paige, Meisel and Colder2021).
Additionally, parental monitoring, peer connectedness, and school engagement did not predict changes in general externalizing symptoms across early adolescence to early adulthood. This diverges from past research that has reported prospective associations between healthy social adjustment and decreases in adolescent problem behavior (Jacobson & Crockett, Reference Jacobson and Crockett2000; Prinstein & La Greca, Reference Prinstein and La Greca2004). Here again, the lack of significant effects in the model predicting general externalizing symptoms increases confidence in the notion that the direction of this association flows from externalizing symptoms to poor social adjustment. This supports findings by Burke et al. (Reference Burke, Pardini and Loeber2008) who investigated bidirectional relationships between parenting and child behaviors, and concluded that evidence for child problem behavior influencing parenting behaviors was stronger than evidence for parenting behaviors predicting child behaviors.
Finally, prospective associations between healthy social adaptation and specific alcohol use and aggression was illuminating. Indeed, adolescents who reported high levels of school engagement prospectively decreased in their levels of specific aggression across time, corroborating past work in this area (Loukas et al., Reference Loukas, Roalson and Herrera2010). Additionally, strong connections to peers predicted high levels of specific alcohol use, lending further support to our interpretation of this factor as healthy and normal trajectories of adolescent drinking. Indeed, a 2023 systematic review and meta-analysis of peer connectedness and substance use in adolescence reported that alcohol use was more likely than other substances to be associated with peer connectedness (Cole et al., Reference Cole, Richmond-Rakerd, Bierce, Norotsky, Peiris and Hussong2023). Contrary to expectation, social adaptation did not predict changes in specific drug use. Analyses testing prospective effects on specific drug use were limited to only two cross-lags, and the stability of specific drug use across time was high.
Limitations
Findings from the current study should be interpreted within the context of certain limitations. First, a large proportion of the sample was male and non-Hispanic White; thus, results may not generalize to samples that are more diverse with respect to sex and race. Additionally, the MLS is a high-risk sample containing a high proportion of children of parents with an AUD. This allowed us to test study hypotheses in a sample of adolescents at risk for high levels of alcohol use, drug use, externalizing problems, and task-general executive functioning deficits; on the other hand, results may not generalize to community samples or typically developing adolescents.
Second, adolescent provided self-report on externalizing symptoms and alcohol and drug use, and also peer connectedness. Reliance on a single reporter may have inflated assocaitions with peer connectedness and potentially deflated relations with other variables from different measurement domains. Still, adolescents’ own perceptions of their ability to make and maintain friendships are likely important in understanding effects of social adjustment on executive functioning development. Future work in this area may elect to use multiple informants when measuring adolescents’ problem behavior, substance use, and social adjustment and examining their impact on executive functioning development.
Relatedly, third, our measure of parental monitoring was limited in several ways. Namely, parental monitoring may not be the ideal measure to utilize when considering the ways in which parents support adolescent cognitive development. Indeed, recent work suggests that measures which more closely assess aspects of parent-adolescent relationship quality (i.e., acceptance and rejection) may shape trajectories of adolescent executive functioning (Hong et al., Reference Hong, Deater-Deckard, Skinner, Duell, Lansford and Steinberg2024). Additionally, our measure of parental monitoring demonstrated fair reliability estimates at several waves, raising concerns about its psychometric properties. On the other hand, general externalizing symptoms predicted changes in parental monitoring in the expected direction, supporting predictive validity of the measure. Still, future work in this area should aim to utilize measures which more precisely and reliabily assess parent-adolescent relationship quality when exploring the role that parents play in supporting the development of adolescent executive functioning.
Conclusion and future directions
Findings from the current study demonstrate that the problem behavior context indirectly predicts poor efficiency of goal-relevant evidence accumulation, a critical ability that underlies individual differences across executive functioning tasks, through social maladaptation. This study was the first, to our knowledge, to provide strong empirical evidence for the role of social adaptation in supporting task-general executive functioning development throughout adolescence and early adulthood. These findings call for research on adolescence to continue to examine the role of social relationships in supporting healthy executive functioning, and support the importance of interventions that target increasing social skills and are based in schools when treating problem behavior (Leff et al., Reference Leff, Waasdorp, Paskewich, Gullan, Jawad, Macevoy, Feinberg and Power2010). Moreover, our results emphasize the need for research in this area to adopt a developmental psychopathology perspective. Findings did not support the notion that there are unique, neurotoxic effects of adolescent alcohol and drug use. It is critical for the field to account for the hierarchical structure of psychopathology, namely the broader externalizing context, when considering the consequences of adolescent alcohol use.
Finally, findings from the current study demonstrate that the application of the DDM model to behavioral tasks of executive functioning may be a fruitful approach in enhancing reliability and construct validity in the measurement of efficiency of evidence accumulation, a construct which is thought to underlie executive functioning (Löffler et al., Reference Löffler, Frischkorn, Hagemann, Sadus and Schubert2024; Weigard et al., Reference Weigard, Clark and Sripada2021; Weigard & Sripada, Reference Weigard and Sripada2021). Indeed, drift rate (v SST ) demonstrated strong psychometric properties in the form of temporal stability that is fair by conventional standards (ICC = 0.52), but relatively robust given the average measurement interval of 2–3 years, and predictive validity in its positive, significant associations with peer relationship quality and school engagement. On the other hand, the application of the DDM model is a relatively new venture within clinical psychology, and further work examining whether results differ when utilizing traditional reaction time-based approaches, as well as a broader battery of assessments, to measure executive functioning tasks is warranted.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S095457942400138X.
Funding statement
This research was supported by grants from the National Institute on Alcohol Abuse and Alcoholism (F31 AA028973) awarded to Katie J. Paige and to Mary M. Heitzeg (R01 AA025790), and a grant from the National Institute on Drug Abuse awarded to Alex S. Weigard (K23 DA051561). The content of this study is solely the responsibility of the authors and does not necessarily represent the official views of NIAAA or NIDA.
Competing interests
None.
Open access
