Abstract
There is limited understanding of mechanisms underpinning the relationship between impulsivity and nicotine vaping. This study seeks to address this gap in the literature by examining deviant peer association as a mediator of this relationship. All five available waves of the Adolescent Brain Cognitive Development study were analyzed (N = 11,880). This is a United States general population sample with participants of ages 8-10 at baseline. Generalized structural equation modeling is used to examine the direct effects of impulsivity on vaping risk and the indirect effect impulsivity running through deviant peer association. Results indicate that lower impulse control/higher impulsivity significantly predicts greater log-odds of vaping at follow-up (Coefficient = .115; p < .003). Greater deviant peer association significantly mediates this relationship (Coefficient = .023; p < .001; 95% confidence interval = .012---.033), accounting for about 15% of the direct effect of impulsivity on vaping. Prevention programming should focus on identifying youth with impulsivity issues and reducing their affiliation with deviant peers.
Introduction
The rise in prevalence of vaping over the past decade has led to a strong focus on identifying predictors of the behavior in order to prevent its use, particularly among youth populations (Hammond et al., 2020; Hammond & Reid, 2025; Tehrani et al., 2022). While recent declines in prevalence have been observed in recent years (Hammond & Reid, 2025) an estimated ∼7% of United States high school students reported past 30-day vaping in 2023 and this comprised more than half of all tobacco product use in this sample (Jackson et al., 2025). The behavior is particularly concerning because there is still much that is unknown about the long-term effects of use. However, preliminary evidence suggests that it may be associated with a range of undesirable physical and mental health outcomes (Bonner et al., 2021; Traboulsi et al., 2020). Two risk factors that may be particularly relevant here are deviant peer association and low impulse control (or alternatively high impulsivity). Both have been identified as risk factors for vaping (Conner et al., 2024; Davis et al., 2022; Lanza et al., 2020; Rocheleau et al., 2020). While these may pertinent because of their observed relationships with vaping, they also may have distinct relevance here because of the potential that they work together to produce a causal pathway to vaping among youth. That said, these factors have yet to be examined together in the manner proposed here. The specific role that peers may play as a facilitator of vaping and the salience of cognitive development during the adolescent years may make them particularly relevant for working together to impact risk for vaping. There has yet to be any study which has examined these risk factors for predicting vaping in a cohesive causal framework. A better understanding of mechanisms underpinning the relationship between impulsivity and vaping could allow more effective targeting of at-risk youth and facilitate design and implementation of more effective prevention programming. This necessitates testing for mediation of the impulsivity-vaping relationship to determine whether deviant peer association provides a mechanism in this regard. The present study sought to address these gaps in the extant literature by examining impulsivity as a predictor of vaping and determining whether deviant peer association significantly mediates this relationship among a sample of youth.
Impulse Control and Deviant Peer Association as Predictors of Vaping
Both low impulse control and greater deviant peer association have been identified by prior research as risk factors for vaping and substance use more generally (Conner et al., 2024; Davis et al., 2022; Hanewinkel & Isensee, 2015; Hoffman, 2024; Intravia et al., 2022; Lanza et al., 2020; Morean et al., 2017; Pentz et al., 2015; Rocheleau et al., 2020; Van Ryzin & Dishion, 2014). Impulse control refers to the capacity to stop and consider potential consequences of an action prior to engagement. Young people with low impulse control then may be presented with the opportunity to vape, but not consider the potential health and legal risks associated with the behavior. This may place them at greater risk for engagement. Impulse control is particularly relevant for understanding this behavior among youth. Prior research has indicated that this aspect of cognition usually does not reach full maturity until the mid-20s at least (Spencer-Smith & Anderson, 2009). As such, youth tend to have limited capacity to stop and think before engaging in behaviors like vaping.
Deviant peer association refers to one’s general friendship patterns with other youth involved in antisocial or illegal behavior. Deviant peer association has been observed as a robust risk factor for engagement in a variety of substance use behaviors (Ladis et al., 2021; Wilhoit & Goodnight, 2023; Yoon et al., 2020), including vaping (Hanafin et al., 2021; Lanza et al., 2020; Rocheleau et al., 2020). There are a number of theoretical perspectives that have attempted to frame this association between peer behavior and one’s own behavior. Social learning theory posits that exposure to deviant peers who endorse pro-substance use values and attitudes that individuals are socialized and internalize these attitudes themselves. They are then at increased risk for involvement in substance use behaviors (Akers, 1973).
Alternatively, it has also been posited that the relationship between deviant peer association and one’s own involvement in antisocial behavior is entirely due to selection effects. Such effects refer to individuals who share interests in substance use or other characteristics tending to coalesce into peer groups characterized by these similarities (Gottfredson & Hirschi, 1990). This “birds of a feather” perspective assumes that the relationship between peer and individual behavior then is actually spurious and only due to these selection effects.
A final perspective that is particularly pertinent for substance use focuses on the role that deviant peers play for facilitating access to drugs/alcohol for individuals who may otherwise have restricted access. In this way, deviant peers are a channel through which individuals may obtain their drug of choice. Indeed, prior research has indicated that peers may play such a role in facilitating access to illicit drugs (Coomber et al., 2016; Coomber & Moyle, 2014; Taylor & Potter, 2013). While vaping tobacco is legal in general, 1 youth use is still restricted by law. As such, youth under the age of 21 are not able to legally buy vapes from the store, thus, restricting their access. Deviant peers involved in various law-breaking behaviors may have additional access to vapes through their own channels. These deviant peers may then offer a channel through which youth may be able to obtain vapes, highlighting the potential importance of deviant peer association for the behavior. While deviant peers may offer a risk factor for understanding vaping behavior through this framework, this perspective remains incomplete. Integration of impulse control may offer a more complete understanding the causal pathway to vaping that integrates the latter two theoretical perspectives mentioned here.
As mentioned above, selection effects have demonstrated some utility for understanding the relationship between peer behavior and one’s own behavior. Drug users tend to spend time with peers who similarly use drugs. Social networks tend to be characterized by homophily, that is, shared characteristics of individuals reporting ties within a social network. Impulsivity has been examined as one of these characteristics which selection effects into peer relationships may operate through. Prior research has indicated that youth may indeed select into peer relationships characterized by similarities in impulsivity (Ragan et al., 2023; vanDellen et al., 2017). For example, Ragan et al. (2023) observed evidence of both homophily in social networks based on impulsivity and also that peer impulsivity significantly predicted one’s own involvement in delinquency. This latter finding is particularly pertinent here, as impulsivity is a robust risk factor for offending (Higgins et al., 2013; Jaynes et al., 2022; Wojciechowski, 2021). As such, homophily based on impulsivity seems likely to result in peer group selection that also results in similarities based on involvement in offending behavior, with prior research supporting this assertion (Geeraert et al., 2024; Richmond et al., 2019; Rokven et al., 2016). Individuals with low impulse control may present increased risk for assortment into peer groups characterized by involvement in rule/law-breaking behaviors. It is here that deviant peer association may provide a mechanism that helps to explain the relationship between low impulse control and vaping.
Having friends involved in rule-breaking behavior may provide channels through which youth can obtain e-cigarettes. Selection into such peer groups because of shared low impulse control of the parties involved then may provide one social mechanism through which low impulse control results in increased risk for vaping. Prior research has indicated the importance of particular actors within drug use networks called social supply dealers. These dealers are members of friend groups who have the role of procuring illicit substances not only for their own use, but also to facilitate use by their friends (Coomber et al., 2016; Coomber & Moyle, 2014; Taylor & Potter, 2013). While impulse control has not yet been examined as a factor relevant for social supply dealing specifically, the prior research on homophily based on impulse control indicates the potential for this social mechanism. Beyond the selection and facilitation theories of impulse control, peers and substance use described here, socialization effects may have some relevance also. Wojciechowski (2023) observed that greater deviant peer association also results in attenuated impulse control net of prior levels of impulse control. In this way, youth may select into peer groups based on similarities in impulse control and deviant peers within these groups may facilitate access to e-cigarettes. However, continued association with these peers may also further inhibit healthy cognitive development that may lead to continued low impulse control and chronic elevated risk for vaping.
Low impulse control may lead to selection into deviant peer groups that facilitate access to e-cigarettes and increases vaping risk among youth. While low impulse control would be expected to still be associated with vaping risk apart from this social mechanism, deviant peer association may still provide a mechanism for partially explaining this relationship. In this way, it would be expected that deviant peer association partially mediates the relationship between impulse control and risk for vaping. While low impulse control would still be expected to increase risk for vaping directly, identification of deviant peer association as an additional mechanism through which impulse control may operate to influence vaping. This opens up the potential for design and implementation of a more diverse array of programming to prevent engagement in the behavior focused on addressing this social factor. There remains a dearth of research which has examined this causal pathway to vaping. This study sought to examine the direct relationship between impulsivity and vaping and determine whether deviant peer association mediates this relationship. The present study sought to address these gaps in the literature by answering the following research question by testing these hypotheses:
Research Question: Does deviant peer association mediate the relationship between impulsivity and vaping?
Lower impulse control will be associated with increased risk of vaping at follow-up.
Deviant peer association will significantly mediate the relationship between impulse control and vaping.
Methods
The present study utilized data from several waves of the Adolescent Brain Cognitive Development (ABCD) study. The data used in the present study were aggregated from the first five waves of the ABCD study, though the key dependent and independent variables were measured at Waves 3-5. This facilitated establishment of temporal ordering between the independent, mediating, and dependent variables in analyses. These waves were chosen over earlier waves because of the very small number of participants who reported vaping prior to Wave 5. This precluded the use of these earlier waves of data for these specific analyses. Observation periods for these waves were one-year in length. This dataset comprises data from 11,880 participants aged 8-10 at baseline measurements. This study continues to follow these participants with annual assessments across 21 study sites in the United States. Data collection is ongoing, with plans to continue following participants until young adulthood. Participant data is collected via surveys about psychosocial and demographic factors. Brain scan data to collect data about the biological underpinnings of social exposures, brain development, and mental illness were also collected. The ABCD consortium was formed in 2015 to oversee data collection and aggregation. All minor participants provided assent to participate in the study, with parents/guardians providing informed consent for their children to participate.
Measures
Vaping
The main dependent variable examined in this study was vaping behavior at Wave 5. Participants were asked whether or not they had reported using a nicotine vape since the prior measurements (∼12 month observation period). A binary variable was used to delineate participants who reported any vaping during the prior observation period from those who did not (0 = No; 1 = Yes). While only 3.48% of participants reported past-year vaping in the sample, the large sample addresses concerns over statistical power, as 163 participants reported vaping during the Wave 5 observation period.
Impulse Control
One of the main independent variables examined in this study was impulse control at Wave 3. Impulse control was measured using the UPPS Impulsivity Scale (Whiteside et al., 2005). This instrument asked participants about various dimensions of impulsivity using a series of ordinal items. The lack of premeditation dimension measures were used to assess impulse control, as these constructs were defined analogously as the capacity to stop and consider potential consequences prior to taking action (e.g., I like to stop and think things over before I do them). These individual ordinal items were summed into an index score so that all participants had a single impulse control score at Wave 3. Higher scores on this index scale corresponded to lower impulse control.
Deviant Peer Association
The other independent variable examined in these analyses was deviant peer association at Wave 4. This construct was measured using the Peer Behavior Profile instrument (Bingham et al., 1995). This instrument measured deviant peer association using three ordinal items that asked participants about the level of involvement in rule-breaking behavior by their friends. The following domains were assessed: skipping school, being suspended from school, shoplifting. Participants were asked how many of their friends were involved in these behaviors using an ordinal scale. An index score was computed from these individual ordinal scores so that every participant had one deviant peer association score at Wave 4. It should be noted that the obtained Cronbach’s alpha score for these items did not quite reach adequate levels (Alpha = .631).
Control Variables
Several control variables were also included in these analyses. Gender was included as a control variable because prior research has indicated that there exist gender disparities in vaping risk (Tehrani et al., 2022). The original coding of the baseline gender variable had three categories: boy, girl, non-binary. However, with only 3 three participants identifying as non-binary, there was no capacity for analyzing these participants in a meaningful way. 2 As such, the non-binary participants were dropped from analyses, leaving a binary variable delineating boy and girl participants (0 = Boy; 1 = Girl).
Another set of control variables included in these analyses assessed race/ethnicity, as past research has indicated that there exist racial/ethnic differences in vaping risk (Cambron, 2023). Race was measured at baseline using a nominal variable with three categories: White, Black, and Other Race. Dummy variables were coded from this original nominal variable which delineated participants in a given race category from all other participants (e.g., 1 = Black; 0 = All other participants). The dummy variable corresponding to White participants was then omitted from analyses as the reference group. Latinx ethnicity was also assessed at baseline and this binary variable which delineated Latinx and Non-Latinx participants was included in analyses as a control variable as well (0 = Non-Latinx; 1 = Latinx).
Parental annual income was also included in these analyses as a control variable, as prior research has indicated that risk for vaping may be stratified by social class (Adebisi et al., 2024). Parental annual income was measured at baseline using an ordinal variable, with higher scores corresponding to greater parental annual income.
Age at Wave 5 was also included as a control variable, as risk for vaping may be elevated among older adolescents compared to younger participants (Chapman & Wu, 2014). Age was measured at Wave 5 in single-year intervals.
Traumatic stress exposure was controlled for in these analyses because this has also been identified as a risk factor related to vaping risk (Martinasek et al., 2021). The Kiddie Schedule for Affective Disorders and Schizophrenia post-traumatic stress disorder module was used to assess traumatic stress exposure at Wave 3 (Puig-Antich & Ryan, 1986). 3 This instrument asked participants about their experiences with various traumatic stressors during the Wave 3 observation period (e.g., Witnessed or caught in a natural disaster that caused significant property damage or personal injury). A count variable was used to provide overall traumatic stress scores for all participants. Higher scores corresponded to participants reporting having experienced a greater variety of traumatic stressors during the prior observation period.
Parental monitoring was also controlled for in analyses because past research has indicated that such monitoring of youth behaviors can be an important protective factor against vaping (Szoko et al., 2021). The Parental Monitoring Survey was administered to youth at Wave 4 to assess this construct (Chilcoat & Anthony, 1996). This scale asked participants about the degree of monitoring of their behavior that they experience from their parents/guardians using a series of ordinal items (e.g., How often do your parents/guardians know where you are?). A mean score was then computed so that all participants had a single parental monitoring score at Wave 4.
The final set of control variables entailed measurement of prior levels of key constructs to control for continuity of the effects of these variables on vaping. Deviant peer association and vaping at Wave 3 were included as control variables in these analyses. These variables were measured analogously to their measurement as key dependent and independent variables as described above (mean score for deviant peer association; binary variable measuring vaping).
Analytic Strategy
The present study utilized generalized structural equation modeling (GSEM) to examine the relevance of impulse control and deviant peer association on vaping risk. GSEM was chosen because of its capacity to breakdown the individual segments of mediation relationships and to be extended so that the statistical significance of such mediation effects could be tested. The generalized form of this method was used because of the binary nature of the vaping dependent variable. Logistic regression was then used within the GSEM framework. Coefficients were interpreted as the predicted differences in the log-odds of vaping given a one-unit increase in an independent variable of interest, net of all other variables. The use of GSEM relaxes the model assumption of multivariate normality. However, the model still needs adequate sample size and a lack of outliers. Both of these assumptions were met. Two models were estimated. Model 1 examined the direct effect of impulse control at Wave 3 on vaping risk at Wave 5 net of all control covariates. Model 2 then included the hypothesized mediating pathway running through Wave 4 deviant peer association, again, net of all control covariates.
Missing data was managed using listwise deletion, as more advanced forms of missing data management were incompatible with GSEM in Stata. The use of multiple imputation to manage missing data would be optimal here, but the platform was incompatible with this method when using the gsem function. That said, traditional structural equation modeling methods were compatible with full-information maximum likelihood estimation. This is not imputation, but rather using all of the available data to construct a covariance matrix that allows for valid results that are consistent with multiple imputation results as well. This linear form of the method was used as a sensitivity analysis despite the binary nature of the dependent variable to determine the degree of confidence that should be had in the findings of interest and these findings are also described in the Results section. Similarly, traditional structural equation modeling model fit indices are not applicable to GSEM models. Stata/MP 16 is also limited in that it cannot provide alternative model fit indices for GSEM models.
The GSEM method has the capacity to be extended to determine whether mediating pathways of interest actually constitute statistically significant mediation of a direct relationship of interest. This necessitates the use of standard errors to compute p-values. While the delta method can be used to compute these standard errors, this can result in non-normally distribution of these standard errors which can then bias estimates of statistical significance. For this reason, the Preacher and Hayes (2008) method of bootstrap resampling was used to compute unbiased standard errors which were then used to determine statistical significance. A total of 500 bootstrap resamples were carried out and pooled to provide standard error scores that adhered to the standard normal distribution and provided robust estimates. The model assumption of temporal precedence was met. The assumption of no omitted variable bias was also assumed to be met through statistical control. The assumption of measurement reliability across waves was met for all variables other than the deviant peer association measure which just missed the threshold for internal consistency.
Results
Descriptive Statistics
aPercentages do not add up to 100% because of some overlap between categories due to multiracial participants.
Generalized Structural Equation Logistic Regression Modeling of Deviant Peer Association on Log-Odds of Vaping at Wave 5 and Proposed Mediating Pathways (N =
Sensitivity analyses were carried out to determine whether the use of full-information maximum likelihood estimation as a more robust method for managing missing data yielded consistent findings when examining the hypothesized relationships using linear structural equation modeling. Findings were generally analogous to those of the main analyses. Deviant peer association accounted for about 18% of the relationship between impulse control and vaping risk and the mediation effect was statistically significant. The consistency of these findings with those of the main analyses provide indication of the robustness of these results. Model fit was also assessed for both of these additional models estimated (Model 1 root mean squared error = .000; Model 2 root mean squared error = .000).
Discussion
This study provided an examination of the social and psychological factors underpinning risk for vaping behaviors among a sample of youth in a single causal pathway. Lower levels of impulse control were associated with increased risk for vaping at follow-up. Greater levels of deviant peer association were found to significantly mediate this relationship, as lower impulse control was associated with greater deviant peer association and greater deviant peer association predicted increased risk for vaping at follow-up. These relationships were robust to control for previous levels of deviant peer association also, indicating additional evidence of selection effects and the potential that deviant peers provided channels for obtaining e-cigarettes. There are a number of important implications of these findings for the treatment and prevention of vaping among youth populations.
Findings from this study indicated that low impulse control was associated with increased risk for vaping, a result that was consistent with prior research on the topic (Conner et al., 2024; Davis et al., 2022). While deviant peer association did significantly mediate this relationship, a large proportion of this effect remained even after inclusion of the mediator in the model and the direct effect remained statistically significant also. This indicates that impulse control remains an important factor to address to reduce vaping risk and/or that there also exist additional mechanisms that underpin this the relationship between impulse control and vaping that should be examined in future research. In terms of leveraging impulse control itself through programming, message framing may provide one tool that may have utility in this regard. Message framing involves exposure to the risks involved with health risks behaviors, generally provided by medical authority figures. Prior research has indicated that message framing has utility for reducing risk for engagement in a variety of risky health behaviors (Gallagher and Updegraff, 2012; Rothman et al., 1993; Salovey et al., 2002), including vaping outcomes (Wu et al., 2024; Zhao et al., 2023). Advances in technology have also facilitated a greater variety of modalities for delivering this type of programming in novel ways that may be particularly salient for youth populations. While various theoretical frameworks have guided these interventions for addressing vaping, the findings of the present study indicate that leveraging programming that helps youth to stop and consider the potential consequences and risks involved with vaping may be crucial here. Programming that integrates the healthy development of impulse control could be particularly relevant in this regard, as educating youth about the risks of vaping while fostering skills to stop and think when actually presented with the opportunity to do so may have the capacity to impact vaping in this population. However, this remains speculative and in need of additional research to better understand how such programming may be designed and implemented in the most effective manner.
Findings of low impulse control increasing risk for vaping must also be considered within developmental context. The dual systems model focuses on in-part on impulse control development during adolescence, indicating that impulse control generally develops at a slow and steady pace across this life-course stage (Steinberg, 2010; Steinberg et al., 2008). The prefrontal cortex is responsible for coordinating communication across brain regions and is one of the final brain regions to reach full maturity, generally not full maturing until the mid-20s (Spencer-Smith & Anderson, 2009). This indicates that this may be an adolescent-specific problem that may be alleviated to some degree upon entering adulthood. However, early onset of tobacco use is also associated with increased risk for chronic issues with the behavior (Sharapova et al., 2020; Vega & Gil, 2005) indicating a need for early intervention to address this issue before it spawns a chronic abuse problem.
Structuring such programs in school settings may provide a context for reaching a universal audience. However, the use of such programming in an intervention setting through diversion programming may also be useful. Rather than subjecting youth caught vaping to juvenile justice sanctions, diverting them into community-based programming may be more beneficial and without the iatrogenic effects of system involvement (Gatti et al., 2009; Rowan et al., 2023). These issues should be considered as design and implementation factors moving forward.
The other key finding of this study was that deviant peer association significantly mediated the relationship between low impulse control and vaping risk, accounting for about 15% of this relationship. This indicates that youth with impulse control issues should be prioritized for programming that addresses their peer relationships in order to reduce risk for vaping. From the perspective of this study, this may be interrupting these relationships so that channels to obtain vapes may be cut off. Such an endeavor may be difficult, as policing the peer relationships of youth is likely beyond the scope of traditional interventions. However, potentially educating parents of youth with impulse control issues about the role that deviant peers play for facilitating access to e-cigarettes may provide one fruitful path for doing so. While monitoring the peer relationships of youth is likely beyond the purview of traditional programming staff, such behavior is very much within the authority of parents. Clarifying the predicted role of peers for providing access to e-cigarettes for parents may provide the needed information about why it is so important for parents to be knowledgeable about the friends that their children spend time with and to interrupt potentially problematic relationships when appropriate. Pairing this with additional education about the potential health risks involved with vaping for the parents may be one potential means of getting parents more involved in this important dimension of their offspring’s development. Doing so may have the capacity to reduce risk for vaping among this population.
Again, consideration of the context in which such programming is provided is necessary as well. Schools may provide the context for identifying youth who are impulsive or at-risk for vaping or deviant peer association, but actually connecting youth to programming can be a hurdle. Outside of diversion programs like those mentioned above, there is no way to mandate this type of programming either. Potentially providing incentives for youth to connect with programming could have utility (Heinrichs, 2006; Parkinson et al., 2019), but again, actual determination of whether such programming is helpful is another question entirely that should be answered before any scaling up of incentives. Future research should seek to pilot study programming that pairs message framing with additional parent training like this.
It should also be noted that impulsivity was associated with increased deviant peer association scores, providing support for the existence of selection effects in the data. This is consistent with prior research on this topic indicating that “birds of a feather” flock together (Weerman & Bijleveld, 2007; Wojciechowski & Lawrence, 2024). Shared characteristics and homophily in social networks mentioned earlier likely explain a significant portion of these selection effects. There also exist a number of neurobiological mechanisms that may precipitate such selection effects (Fox, 2017), indicating that affiliation with deviant peer groups may be in-part caused by genetics. Gene-environment correlational research has begun to explore these selection effects, with some research indicating that between 30%-50% of the variation in peer group selection may be due to genetic factors (Kendler et al., 2007). While this is still an area in great need of additional study, it does highlight the potential for biological mechanisms being relevant in this area that was previously considered complete domain of sociologists. Exploration of these processes was beyond the scope of this study, but this highlights a need for continued research in this area.
Gender was also observed to influence vaping risk, with girls reporting greater risk for vaping than boys in these analyses. This is particularly interesting within the context of the data itself. A recent meta-analysis indicated that United States boys reported greater prevalence of vaping, but that girls reported greater prevalence of flavored vaping (DeVito et al., 2025). This raises the question of whether these findings were inconsistent with the general proclivity of boys to vape and use substances at higher prevalence or potentially vaping of flavored substances existed at higher levels in these data. While further examination of these data issues was beyond the scope of this study, it does highlight a need to re-examine these processes in other datasets to determine robustness, particularly data that involves flavored vaping devise specifically.
While the present study provided an important examination of the interplay between social and psychological predictors of vaping, there remain several noteworthy limitations. The first of these limitations pertains to the deviant peer association measure used in analyses. This variable provided a general score of peer involvement in rule-breaking behavior, but it may have lacked in terms of content validity. Only three rule-breaking behaviors were included in the measure: school suspension, shoplifting, and skipping school. While all three of these measures certainly capture deviant peer association as a concept, there are clearly a range of other deviant behaviors by peers that are not assessed with this measure. This may be particularly pertinent as it relates to potential substance use behaviors by peers since vaping was the outcome of interest here. However, given the low prevalence of vaping in the sample overall when assessed at Wave 5 (∼1%), substance use was going to be limited in its measurement among peers as well. This is likely due to the sample still being relatively young at this point, with few youth involved in substance use and antisocial behaviors at this point. For this reason, the more general measure used in analyses may have been most fit to capture general deviant behavior among peers. However, this still indicates the need for continued research on these relationships using measures of peer behavior that encompass a wider variety of behavioral dimensions.
Another limitation of this study pertains to the limited capacity of missing data management in the study. The GSEM command in Stata is not compatible with more advanced forms of missing data management like multiple imputation and maximum likelihood estimation. This led to only listwise deletion being an available option and obviously this is not ideal for ensuring that all relationships observed in the study were valid. If differential attrition was present on one or more variables, then this may have impacted the validity of findings. Multiple imputation specifically could have addressed this issue if available by providing non-biased estimation of data in areas where it was missing based on regressing the missing data on predictors of missingness. Full-information maximum likelihood also could have aided int his regard and provided estimates consistent with multiple imputation, but the method’s utility in a nonlinear context is much more restricted. While traditional linear structural equation modeling paired with full-information maximum likelihood estimation was used as a sensitivity analysis, this was still an imperfect approach because of the binary dependent variable. This indicates the need for additional research on these relationships using more advanced forms of missing data management to determine the robustness of these findings.
Another limitation pertains to the possibility of reciprocal relationships across waves. Examination of reciprocal relationships was beyond the scope of this study and outside the bounds of the theoretical model tested, but there should be concern about the potential of key variables influencing one another across measurement waves. There should also be concern about the potential for omitted variable bias. While statistical control of all of the included constructs should address this concern, there always remains the potential for spurious relationships due to certain variables not accounted for. Generalizability and attrition should also be considered as key concerns. While the ABCD sample is a diverse and rigorously collected dataset, it is not a true representative sample of the United States general population of youth. Results may lack generalizability to other cultural contexts and outside of the age range examined for this reason. Attrition may also have impacted results, as over 50% of the original sample had missing data at Wave 5 for at least one variable included in analyses. There should also be concerns regarding the potential for measurement error in the data, particularly error that may arise from youth self-report. This may may be a particular concern for the impulsivity and deviant peer association measures due to inaccurate reporting and projection of one’s own behavior onto their peers, respectively. It should also be noted that not including delinquency as a potential confounder of relationships is another limitation. While these data do exist in the ABCD-Social Development sub-study data, this only comprised about 2000 participants at baseline, so these data could not be included for the full sample. These issues highlight a need for additional examination of these processes using data without these limitations.
Footnotes
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.