Anxiety as a Moderator of the Relationship Between Deviant Peer Association and Substance Use: Does the Specific Dimension of Anxiety Matter?

Deviant Peer Association as a Risk Factor for Substance Use

As noted above, deviant peer association has been identified as a robust risk factor predicting substance use (Kim-Spoon et al., 2019; Price et al., 2019; Wojciechowski, 2018). Perhaps the most relevant criminological theoretical perspective aiding to explain this relationship is Akers’ (1973) social learning theory. This framework posits that exposure to individuals who engage in deviant behavior, like substance use, provide models through which individuals learn the motivations, rationalization, and techniques pertaining to engagement in deviant behavior. The theory is general in the sense that prosocial behaviors are transmitted through a similar learning process. Individuals may provide definitions that are favorable or unfavorable toward a specific behavior. Exposure to more definitions that are favorable toward a behavior then should increase the likelihood that an individual will engage in the given behavior themselves. For substance use, this may entail exposure to peers use drugs or alcohol themselves and present models for engagement in the behavior for a given target individual. There also exist other mechanisms by which peers may influence behavior in this regard. Processes pertaining to reinforcement or punishment of a behavior may also increase or decrease the likelihood of future engagement in said behaviors. Reinforcement refers to any action or potential action that may be taken by other individuals to increase the likelihood of future engagement in a behavior. For example, peers may praise or encourage a target individuals first engagement in substance use and this may lead the target individual to feel rewarded for engagement in the behavior, thus, increasing the likelihood that they will do so again in the future. Punishment is the opposite of this process, as it refers to actions or potential actions that may reduce the likelihood of future engagement in a behavior. Prosocial peers may indicate a reticence to maintain relationships with an individual upon discovering that they have been engaging in substance use. The potential for losing these friendship ties then may act as a threatened punishment that should result in lower risk for the individual engaging in substance use in the future. It is through these highlighted mechanisms that peer relationships are relevant for understanding behavior and why deviant peer ties may specifically be important for engaging in antisocial behavior like substance use. Prior research has indicated support for the relevance of these mechanisms for predicting substance use (Pratt et al., 2010; Kruis et al., 2020). For example, Pratt et al., 2010 provide a meta-analysis indicating that all mechanisms of the theory are relevant for predicting behavior, though differential association and modeling/imitation seem most relevant. However, there remains continued need to understand how individual characteristics may amplify or attenuate the salience of deviant peer association for influencing engagement in substance use. One key factor here may be anxiety.

Anxiety Dimensions as Moderators of the Relationship Between Deviant Peer Association and Substance Use

The persistent unease and worry that characterizes the mental condition of anxiety may present an important psychological factor for understanding the relationship between deviant peer association and substance use. It might be expected that adolescents who report greater anxiety may have exacerbated response to potential social punishments from deviant peers. Individuals may perceive that deviant friends may have limited desire to maintain friendships if they themselves to not engage in antisocial behavior like substance use. For individuals high in anxiety, oversensitivity to such concerns may then exacerbate the risk for substance use as they persistently dwell on this worry. While this would make sense, these relationships as they pertain to general anxiety remain understudied. In the case of general anxiety, prior research has even found that anxiety may moderate this relationship in the opposite direction. Rodriguez et al. (2019) found that peer substance use did not influence the substance use behaviors of target adolescents who were high in anxiety. Rather, peer substance use only influenced the substance use of adolescents who reported low anxiety. These relationships also must be understood within the context that some research has suggested that anxiety may also act as a protective factor against substance use (Kaplow et al., 2001; Myers et al., 2003). For example, Kaplow et al. (2001) found that while generalized anxiety was a risk factor for alcohol use, symptoms of separation anxiety was associated with lower risk. This study highlights the fact that, while anxiety may represent a broad range of related symptoms, there do exist distinct dimensions of anxiety and the specific nature of the anxiety may matter for how it impacts the relationships of interest; like that of the relationship between deviant peer association and substance use. It may be that some dimensions of anxiety act as a protective factor against the influence of peers on substance use, whereas other dimensions act in the manner posited above leading to greater risk for substance use. In a situation like this, examining general anxiety as a moderator may be obscuring the moderation effects of more specific dimensions that could act as a moderator in the manner proposed above. This necessitates delineation of these specific dimensions of anxiety in order to determine if certain forms of anxiety protect against peer influence in this regard and if other dimensions exacerbate the effects of peer influence.

Examinations of the roles that specific dimensions of anxiety play for moderating the relationship between deviant peer association and substance use has also been limited. The main branch of research in this area has focused on social anxiety as a moderator specifically. This makes sense considering that social anxiety should directly correspond to the reactions that individuals may have to peers attempting to influence their behavior. If individuals are concerned about how they might interact with peers, this may also lead them to engage in substance use as a means of coping or alleviating perceived social awkwardness. Having greater levels of deviant peer association may then facilitate substance use and make this behavior a realistic option for reducing this form of anxiety during social situations. Findings from studies focused on social anxiety in this role generally indicate that social anxiety does moderate the relationship between deviant peers and substance use in the expected manner, with high levels of both social anxiety and deviant peer association predicting elevated substance use risk (Anderson et al., 2011; Cohen & Prinstein, 2006). For example, Cohen & Prinstein (2006) found that greater social anxiety resulted in greater conformity to peer norms and that this manifested in greater substance use. There may also be gendered effects at play here, as Anderson et al. (2011) found a stronger peer effect on binge drinking specifically for boys who were high in social anxiety. This provides indication that while general anxiety may generate a protective effect against substance use and the influence that peers may exert on substance use, generalized measurement may obscure the increased risk for substance use associated with social anxiety specifically. Past research has also indicated that social anxiety may have a direct effect resulting in increased substance use risk also (Mueller et al., 2021; Villarosa-Hurlocker et al., 2019). For example, Mueller et al., (2021) found that social anxiety increased risk for marijuana use problems and that this relationship was stronger for males. While prior research provides some indication of this, there remains a dearth of research on the other specific forms of anxiety of interest here: worry/oversensitivity anxiety and physiological anxiety.

Worry/oversensitivity anxiety refers to symptoms characterized by vague obsessive concerns about may be nonspecific in nature (e.g., “what if something bad happens?”). This is delineated from social anxiety in the sense that these concerns are not distinctly focused on anxiety driven by concerns regarding interpersonal interaction. Physiological anxiety here refers to the actual physical symptoms that may accompany physical or tangible symptoms of anxiety (e.g., sweating palms, waking up scared, bad dreams, etc.). In the case of worry/oversensitivity anxiety, overwhelming concerns regarding potential consequences or issues stemming from substance use may lead individuals to have lower risk for engagement in substance use behaviors. Such worries may be related to getting into trouble for engagement in illicit behavior, as this may pertain to a range of potential consequences (ex: going to jail, being suspended from school, being grounded by parents, etc.). Because these concerns would not be directly related to the potential for social awkwardness or the desire to feel less inhibition in social situations, this may actually lead individuals to be a protective factor against substance use. In the context of deviant peers, this may also play a protective role, as individuals high in worry/oversensitivity anxiety may be buffered against the effects of deviant peers on their substance use risk. In the case of physiological anxiety, potential moderation effects pertaining to deviant peer association and substance use seem less clear. While individuals may attempt to use substances as a means of coping with these symptoms in a manner consistent with the self-medication hypothesis (Khantzian, 1987, 1997), it is unclear that this would relate to deviant peer association. It would seem to depend on if these physiological symptoms are connected in any way to either of the other forms of anxiety. These physiological symptoms may then be manifestations of either worry/oversensitivity anxiety or social concerns/concentration anxiety. If this is the case, then it would seem that the other forms of anxiety may be more relevant for moderating the relationship between deviant peer association and substance use. Physiological anxiety then should not actually act as a means of moderating this relationship, as this should just be a manifestation of anxiety that corresponds to other types. In this way, it would be expected that physiological anxiety should not significantly moderate the relationship between deviant peer association and substance use.

The moderating role of anxiety for understanding the relationship between deviant peer association and substance use has been examined by prior research. However, findings regarding the nature of this relationship have been mixed (Cohen & Prinstein, 2006; Rodriguez et al., 2019). For example, Cohen and Prinstein (2006) found that youth high in social anxiety were susceptible to influence of both high and low status peers, whereas those low in social anxiety were only susceptible to high status peers; providing indication that high anxiety may increase susceptibility to peer influence. Rodriguez et al. (2019) found the opposite, as youth high in anxiety were not as vulnerable to peer influence on substance use compared to youth with low levels of anxiety. This indicates the need for continued examination of specific anxiety dimensions as moderators of the relationship between deviant peer association and substance use. It may be that general anxiety measures may be too coarse, as worry/oversensitivity anxiety and physiological anxiety may be masking or washing out amplification effects of social concerns/concentration anxiety. The present study sought to add to the extant literature by testing the following hypotheses:

Data

Data utilized in analyses were from the first two waves of the Pathways to Desistance study. This dataset was comprised of the responses of 1354 justice-involved youth who then provided data for the following 7 years after being adjudicated for a serious offense prior to baseline measurements; resulting in a total of 11 waves of data. Serious offenses which met inclusion criteria consisted of all felony offenses, as well as sexual assault and weapons-related misdemeanors. Offenses which qualified participants for inclusion in the study had to have been committed when participants were between the ages of 14 and 17 in order to meet inclusion criteria. The first two waves were utilized because they had the least missing data and provided the most parsimonious means of testing hypotheses while also establishing temporal ordering between the independent and dependent variables as a necessary condition for inferring causality. Participants were recruited from study sites located in Maricopa County, Arizona and Philadelphia, Pennsylvania. The proportion of male drug offenders included in the sample was capped at 15% in order to maintain heterogeneity in these baseline characteristics. Of all qualified youth approached regarding their interest in the study, 20% declined the opportunity to participate. Of the baseline sample, 93.4% were retained and provided data for wave two measurements. There was a 6-month observation period between baseline measurements and wave two measurements.

Data used in this study were collected via participant self-report. Interviews were conducted in locations that were convenient for participants (e.g., libraries, criminal justice facilities, participants’ homes, etc.). The research team provided each participant with a laptop computer during interviews. Participants then manually input responses to verbal prompts from the research team into the laptop. This method of data collection was chosen in order to maximize confidentiality in reporting on sensitive topics.

Measures

Deviant Peer Influence

One of the key independent variables examined in analyses was deviant peer influence at baseline. This construct was measured using a scale adapted from the Rochester Youth Development Study (Thornberry et al. 1994). This scale was a measure of the general number of peers who participants reported attempt to influence them to engage in antisocial behavior. Seven ordinal items were used to assess how many friends that participants reported having that influenced them to engage in seven different antisocial behaviors. A mean score was then computed from these individual items so that each participant had a single deviant peer association score at baseline. The following items were utilized for the deviant peer association measure: How many of your friends have suggested that you should go out drinking with them?; How many of your friends have suggested or claimed that you have to get drunk to have a good time?; How many of your friends have suggested or claimed that you have to be high on drugs to have a good time?; How many of your friends have suggested that you should sell drugs?; How many of your friends have suggested that you should steal something?; How many of your friends have suggested that you should hit or beat someone up?; How many of your friends have suggested that you should carry a weapon? Cronbach’s alpha indicated good reliability for this measure (.89).

Anxiety

The other set of key independent variables included in analyses focused on different forms of anxiety that participants reported experiencing. Three different forms of anxiety were measured: physiological anxiety, worry/oversensitivity, and social concerns/concentration. The physiological anxiety items focused on somatic manifestations of anxiety. The worry/oversensitivity items focused on internal anxiety about fears related to being hurt or emotionally isolated that may sometimes be vague and difficult to express clearly. Finally, the social concerns/concentration items focused on feelings of anxiety centered distracting thoughts and fears that have a social and/or interpersonal focus. The Revised Children’s Manifest Anxiety Scale was used to assess each type of anxiety at baseline. This consisted of 37 items that asked participants to report whether or not they experienced individual symptoms of anxiety using binary measures. These binary measures allowed participants to report whether or not each symptom was currently descriptive of their own feelings or actions related to the experience of anxiety. Count scores were then computed for each type of anxiety, providing a count of the number of symptoms that participants endorsed for each anxiety dimension. This then resulted in three anxiety scores for each participant at baseline that were included in analyses. The following items were used to assess physiological anxiety: I often have trouble making up my mind; Often I have trouble getting my breath; I get mad easily; It is hard for me to get to sleep at night; Often I feel sick in my stomach; My hands feel sweaty; I am tired a lot; I have bad dreams; I wake up scared some of the time; I wiggle in my seat a lot. The following items were used to assess worry/oversensitivity anxiety: I get nervous when things do not go the right way for me; I worry a lot of the time; I am afraid of a lot of things; I worry about what my parent will say to me; I worry about what other people think about me; My feelings get easily hurt; I worry about what is going to happen; My feelings get easily hurt when I am fussed at; I worry when I go to bed at night; I am nervous; I often worry about something bad happening to me. The following items were used to assess social concerns/concentration anxiety: Others seem to do things easier than I can; I feel that others do not like the way I do things; I feel alone even when there are people with me; Other people are happier than I; I feel someone will tell me I do things the wrong way; It is hard for me to keep my mind on my school work; A lot of people are against me. Cronbach’s alpha scores for these variables indicated adequate reliability for worry/oversensitivity anxiety (.81), but somewhat suboptimum reliability for the physiological anxiety (.66) and social concerns/concentration anxiety measures (.69).

Control Variables

Several control variables were also included in analyses so that risk for bias in estimation may be mitigated. The first of these variables was gender, as prior research has indicated greater differences in substance use risk (McHugh et al., 2018). Gender was assessed at baseline using a binary variable that delineated male and female participants (0=Male; 1=Female).

Race was also included as a control variable, as prior research has indicated that White youth may present greater substance use risk than youth identifying in other Non-White racial groups (Vasilenko et al., 2017). Race was assessed at baseline using a four-category nominal variable with the following racial identification options: Black, Hispanic, White, and Other Race. A series of four dummy variables was then coded, with each dummy variable delineating participants in a single race category from all other participants and was repeated for each race category (e.g., 1=Black; 0=All other participants. The dummy variable corresponding to White participants was then omitted from analyses in order to provide a reference group for interpreting coefficients for the other race variables in comparison with.

Socioeconomic status (SES) was another control variable included in analyses, as prior research has indicated that substance use risk is stratified by SES (Vilsaint et al., 2019). SES was measured at baseline utilizing Hollingshead’s (1957) two-factor index of social position. This instrument measures SES as a weighted score comprised of participants’ parents’ educational attainment and occupational prestige scores. If both parents were available to provide data, then a mean score was computed so that all participants had a single baseline SES score.

Another control variable included in analyses was exposure to violence. This is because past research has indicated that exposure to violence is a risk factor for substance use (Mitchell et al., 2007). This construct was measured at baseline using the Exposure to Violence Inventory (Selner-Ohagan et al., 1998). This scale assessed whether or not participants had ever experienced different forms of direct victimization and/or witnessed violence prior to baseline. A count score was then generated that provided a measure of the number of different types of exposure to violence that each participant experienced prior to baseline.

Impulse control at baseline was also controlled for, as prior research has indicated that low impulse control is a risk factor predicting substance use (Davis et al., 2017). This was measured using the Weinberger Adjustment Inventory (Weinberger et al., 1989). This scale was comprised of a series of ordinal items that asked participants to rate the degree to which a set of statements related to impulse control was true of their own behavior or attitudes (e.g., I say the first thing that comes into my mind without thinking enough about it). Seven of the eight individual items were reverse coded so that higher scores predicted higher impulse control. A mean score was also computed so that all participants had a single impulse control score at baseline. Cronbach’s alpha indicated adequate reliability for this measure (.76).

Age was also included as a control variable, as prior research has indicated that substance use risk may be age-graded (Vasilenko et al., 2017). Age was measured at baseline in single-digit year intervals.

Time spent in secured facilities with no community access was also controlled for (e.g., jails, prisons, psychiatric hospitals), as spending more time without community access during the wave two observation period may have impacted one’s capacity to engage in substance use. This variable was operationalized as a proportion ranging from 0 to 100, with higher scores indicating more time spent in secured facilities during the wave two observation period (e.g., 55 = 55% of the wave two observation period spent in secured facilities).

Wave two observation period length was also controlled for, as longer observation periods offered increased exposure time with which to engage in substance use and this may have impacted rates of use. While all participants’ wave two observation periods were generally 6 months in length, there was some variance in the exact number of days. As such, an indicator of the exact number of days in each participant’s wave two observation period was included in analyses.

Analytic Strategy

Ordered logistic regression was utilized to examine each substance use outcome. This modeling strategy was chosen because of the ordinal measurement of each of the dependent variables. Two models were estimated for each dependent variable. Model 1 for each variable examined the direct effects of the baseline deviant peer association and anxiety measures for predicting substance use. Model 2 for each dependent variable then included an interaction term that modeled the moderating effect of all three anxiety dimensions on the relationship between deviant peer association and substance use. These interaction terms were computed by multiplying each participant’s deviant peer association score by their scores on each anxiety dimension. The deviant peer association and all anxiety measures were also mean-centered for these analyses by subtracting the mean value from individual participants’ values in order to address concerns of multicollinearity in calculating interaction effects. Listwise deletion was used to manage missing data. This missing data management strategy was chosen for several reasons. First, preliminary analyses indicated that there was not a strong correlation between predictors and missingness, with only about 7% of participants from the original sample no longer providing data at wave 2. This 7% missingness was observed fairly evenly across all measures included in analyses, indicating that there was not one specific measure that had significantly more missing data than others. Further, some research has indicated that listwise deletion may actually provide the most accurate estimation of a full dataset for criminal justice populations (Mitchell et al., 2021). Coefficients are described in the form of odds ratios (ORs), indicating the impact that a one-unit increase in an independent variable of interest had on the odds of being in a higher category on the ordinal scale for each substance use outcome. ²