Youth with Co-Occurring Delinquency and Depressive Symptoms: Do They Have Better or Worse Delinquent Outcomes?

Abstract

Delinquent youth often experience depression, but depression’s impact on their future deviance is unclear. Using survey and social network data on a panel of 9th graders (N=8,701; M_age at baseline = 15.6; 48% male; 85% white; 18% eligible for free or reduced-price school lunch) followed throughout high school, this study tested whether depressive symptoms predicted later deviance or deviant peer affiliations among already delinquent youth. A latent class analysis revealed that 4% of respondents showed above-average levels of delinquency but not depressive symptoms, and 3% were above average on both. Compared to the delinquent-only group, the delinquent-depressed group went on to have less deviant friends, and to engage in less deviance themselves. However, peer deviance was not a reliable explanation for the reductions in respondents’ own future deviance. Depressive symptoms thus may play a protective role against continued delinquency and substance use among youth who are already delinquent, but it is not because they reduce deviant peer affiliations.

Introduction

Co-occurring delinquency and depressive symptoms are common among adolescents (Capaldi and Kim 2014). This co-occurrence predicts worse mental health outcomes (Reinke, Eddy, Dishion, and Reid 2012), but it is unclear whether it also predicts worse delinquent outcomes. Given that adolescence is the peak age range for the escalation of delinquency, in large part because of increasing delinquent peer affiliations (Sweeten, Piquero, and Steinberg 2013), it is important to understand whether and how having both problems at once might influence this escalation by influencing peer processes. Theoretically, co-occurring depressive symptoms could steer youth either toward or away from deviant peers (Hussong, Jones, Stein, Baucom, and Boeding 2011), and those altered deviant peer affiliations could in turn either increase or decrease adolescents’ own delinquency. Yet research has not yet determined which of these possibilities is the case. This study tests whether co-occurring delinquency and depressive symptoms in adolescence worsen or improve future delinquency, rates of police contact, and substance use, and, if so, whether they do so through increased or decreased delinquent peer affiliations.

How Peers Might Mediate the Path from Co-Occurring Delinquency and Depressive Symptoms to Deviant Outcomes

The finding that depressive symptoms and delinquency positively co-occur during adolescence is at least three decades old (Capaldi and Kim 2014). There are three broad potential reasons for this co-occurrence (see Wolff and Ollendick 2006 for a discussion). First, under a shared risk model, the two conditions may share root causes such as genetic liabilities. Second, under a failure model, delinquency may cause depression by triggering social rejection. And third, under an acting-out model, stressors may lead to delinquency by triggering negative affective states such as depression, which create internal pressure to act out. Peers play a central role in the second model; they may play one under the third as well, if negative peer relations qualify as a stressor. That is, under these models peer relations may mediate the longitudinal association between delinquency and depressive symptoms.

These peer-based models of delinquency and depression’s co-occurrence focus on the quality of peer relations, rather than on the peers’ actual behavior. Yet studies from network science consistently show that adolescents’ behavior is influenced by their peers’ behavior. For example, peer involvement in a variety of forms of deviance, including delinquency, aggression, violence, and weapon carrying, influenced adolescents’ own involvement in those same behaviors (Sijtsema and Lindenberg 2018). With respect to depression, one study found that adolescents’ depressive symptoms were predicted by their friends’ symptoms (Van Zalk, Kerr, Branje, Stattin, and Meeus 2010). (For null findings on peer influence for depression and violence, see Schaefer, Kornienko, and Fox [2011] and Turanovic and Young [2016]). These findings raise the possibility that friends’ behavior may also contribute to crossover effects between depressive symptoms and delinquency. Consistent with this idea, adolescents’ delinquency is predicted not only by their friends’ delinquency, but also by their friends’ depression (Reynolds and Crea 2015).

Two competing theoretical pathways could explain how co-occurring depressive symptoms may influence delinquent adolescents’ affiliation with deviant peers. First, the symptoms could increase deviant peer affiliation by steering adolescents away from non-deviant peers (Hussong et al. 2011). Under this possibility, the negative self-images that accompany depression lead delinquent-depressed adolescents to reject conventional others, who they perceive to be a source of negative evaluations. Such rejection limits the pool of potential friends to deviant youth, who are appealing, and thus befriended, because they model alternative ways of behaving and developing positive self-attitudes (Damphousse and Kaplan 1998). This perspective assumes that depressed adolescents will be motivated to find peer acceptance and interaction, and will resort to finding these with deviant peers once they have rejected, or have been rejected by, mainstream peers.

Although some scholars have assumed that withdrawing from prosocial peers means that delinquent-depressed adolescents will increase their affiliation with antisocial peers, this may not be the case. A second possibility is that internalizing symptoms could decrease deviant peer affiliation by leading to general social withdrawal (Hussong et al. 2011). That is, the withdrawn behavior that often accompanies depression could reduce adolescents’ interest in befriending deviant peers, even if they have few other friends. This, in turn, may actually protect them against future delinquency. Indeed, socially isolated adolescents are no more delinquent, and may even be less delinquent, than their mainstream peers (Demuth 2004; Haynie 2002). It thus is possible that co-occurring symptoms of depression reduce future involvement in delinquency and substance use by reducing involvement in peer contexts that are likely to promote deviance. Research has not yet fully tested this possibility.

Co-Occurring Delinquency and Depression, Peers, and Deviant Outcomes

Although there are few direct tests of the full pathway from co-occurring delinquency and depression to deviant peer affiliations to delinquent outcomes, many studies have examined how the combination of delinquency or conduct problems and depressive symptoms relates to future adolescent and young adult deviance. Several studies have found that that combination positively predicts future aggression (Kang et al. 2015), criminal offending (Sourander et al. 2007), serious arrest (Copeland, Miller-Johnson, Keeler, Angold, and Costello 2007), and substance use (Maslowsky and Schulenberg 2013). In contrast, though, an equally large number of studies have found that the combination does not predict general rule violation (Fanti and Henrich 2010) or less serious arrests (Copeland et al. 2007), and that co-morbid depression in comparison with “pure” delinquency has no effect, or even a beneficial effect, on substance use (Khoddam, Jackson, and Leventhal 2016). Furthermore, and importantly for theory, there is mixed evidence on the implications of co-occurring depressive symptoms for peer relations. Studies have alternately found that among delinquent youth, depressive symptoms—as well as internalizing symptoms more broadly—have no effect on affiliation with deviant peers (Capaldi and Stoolmiller 1999; Ingoldsby et al. 2006) or actually reduce affiliation with deviant peers (Mason, Hitchings, and Spoth 2008; Scalco et al. 2014).

Only two studies have directly examined whether co-occurring depressive symptoms predict peer deviance, and in turn adolescents’ own deviance, among delinquent youth. One found that the interaction of conduct problems and depressed mood from ages 11–16 negatively predicted substance use at age 18, and that this interaction effect was in part indirect through peer substance use (Mason et al. 2008). The other found that the interaction of conduct problems and internalizing problems at ages 11–12 negatively predicted substance use two years later indirectly through its negative association with peer delinquency (Scalco et al. 2014). These results suggest that depressive symptoms may protect against future deviance among delinquent youth by reducing their exposure to peer deviance.

These studies, while highly informative about the outcomes of comorbid delinquency and depression, leave two gaps in the literature. First, they examined only substance use. As noted above, the co-occurring problems potentially are linked to a wide range of deviant outcomes, so it important to understand whether a peer-linked theoretical pathway holds for outcomes beyond substance use. Because depression is an internalizing problem, it is possible that it triggers the peer-linked pathway more with respect to internalizing forms of deviance (like substance use) than it does with respect to other forms of deviance (like law violation). Second, the two most relevant past studies used interaction terms to represent the co-occurrence of delinquency and depressive symptoms. Although this strategy can tell us whether youth with the two co-occurring problems have unusually positive or negative outcomes, it cannot tell us whether they fare better or worse specifically relative to youth with “pure” delinquency. This is because a significant interaction may simply indicate that delinquent-depressed youth differ from youth who only have depressive symptoms, not necessarily that they differ from youth who are only delinquent. There is some evidence that this is the case: For instance, in one study, post hoc tests of slopes indicated that despite a significant interaction term, depressive symptoms did not predict substance use among the subset of youth with elevated conduct problems (Mason et al. 2008). Interaction terms alone thus cannot tell us whether depressive symptoms influence the outcomes of already delinquent youth.

Recognizing these issues, some recent studies have used latent class analysis (LCA; e.g., Brook et al. 2015). This strategy overcomes the above limitations by identifying distinct groups of adolescents with similar patterns of delinquency and depressive symptoms. Studies using this approach have identified groups of adolescents showing co-occurring delinquency and depressive symptoms, “pure” delinquency, and “pure” depressive symptoms, as well as other groups. These studies have yielded inconsistent findings about the consequences of co-occurrence for peer and behavioral outcomes. For instance, one study found that the few youth who show both high disruptive behavior and high depressive symptoms in childhood and adolescence have very high probabilities of young adult arrest and substance use problems (Reinke et al. 2012). In contrast, another found no differences between children showing pure versus co-occurring internalizing and externalizing problems in early adolescent risky behavior (experimenting with weapons and substance use) or in affiliation with friends involved in risky behavior (Fanti and Heinrich 2010). Adding to the mixed findings, a third study found that adolescents high on both delinquency and depressive symptoms showed elevated aggression, but not delinquency, in adulthood (Diamantopoulou, Verhulst, and Van Der Ende 2011). Although some of these results suggest that co-occurring depressive symptoms worsen delinquent outcomes, the overall pattern is inconclusive.

Current Study

This study tests three sets of hypotheses. First, it tests whether the combination of delinquency and depressive symptoms positively (Hypothesis 1a) or negatively (Hypothesis 1b) predicts peer delinquency, peer police contact, and peer substance use. Under one prominent theoretical pathway, delinquent youth who are also depressed experience more rejection by mainstream peers and are driven to find acceptance from deviant peers. If this is the case, then co-occurring depressive symptoms should positively predict deviant peer affiliations among delinquent youth. In contrast, delinquent youth who are depressed may also be socially withdrawn, and may have few affiliations with either deviant or non-deviant peers. If this is the case, then co-occurring depressive symptoms should negatively predict deviant peer affiliations among delinquent youth.

Second, the study tests whether the combination of delinquency and depressive symptoms positively (Hypothesis 2a) or negatively (Hypothesis 2b) predicts respondents’ own delinquency, police contact, and substance use. Because peer deviance is a key predictor of adolescents’ own deviance, if one of the proposed theoretical pathways is correct, it ultimately should result in increased or decreased levels of future deviance. Because depression is an internalizing problem, it is possible that the strength of the pathway will be different (i.e., greater) for substance use than for delinquency and police contact.

Third, the study tests whether peer delinquency and peer substance use explain any observed associations between co-occurring delinquency and depressive symptoms and future deviance (Hypothesis 3). They should mediate those associations if the proposed pathway is correct.

Finally, to guard against spuriousness, the study accounts for several known correlates of delinquency, internalizing symptoms, and peer relations, namely demographic characteristics, family structure and relations, school grades and attachment, religious attendance, and sensation seeking (cf. Osgood et al. 2013; Siennick et al. 2016; Tucker et al. 2012).

Methods

Data

The data are from PROSPER (PROmoting School-Community-University Partnerships to Enhance Resilience), a place-randomized substance abuse prevention trial in 28 public school districts in rural Pennsylvania and Iowa (Spoth, Greenberg, Bierman, and Redmond 2004; Spoth et al. 2007). PROSPER included school districts that enrolled between 1,300 and 5,200 students, and had student populations with at least 15% of families eligible for free or reduced cost school lunch. PROSPER sampled two successive cohorts of students who were first surveyed in the fall of 6th grade (in 2002 and 2003) with follow-ups conducted each spring from 6th through 12th grade. An average of 10,000 students completed surveys at each wave. The surveys included information on depression starting in 9th grade. This study thus used the last four waves of data collection and began with the 10,785 respondents who participated in the 9th grade survey.¹ This included 82% of the original 6th grade respondents, as well as additional respondents who had moved into the study schools after 6th grade and were recruited into the sample.

The focal predictor was measured in 9th grade and the focal outcomes were measured at the remaining three waves. Respondents thus were selected for inclusion in the study if they had complete data on depressive symptoms and delinquency in 9th grade (average age 15.5) and participated in at least one additional survey during the 10th through 12th grade assessments (average ages 16.6–18.6). Of the 10,327 respondents who met the former criterion, 814 were excluded because they did not complete a 10th through 12th grade survey, and another 812 because they provided no information on the focal dependent or mediating variables in these grades. Attrition analyses revealed that relative to 9th grade respondents who were excluded, respondents in the analytical sample were more likely to be female, white, and living with both biological parents, and they were less likely to receive free school lunch. On average they also had better family relations, attended religious services more often, reported less sensation seeking, and had better school grades and higher school attachment. These differences were modest in size (i.e., an average difference of 10% on dichotomous measures and 0.11 standard deviations on continuous measures). Still, this study’s results may not generalize as well to youth without these characteristics.

The final analytical sample consists of 8,701 respondents contributing 19,405 10th-12th grade observations. These respondents were 48% male and 85% white. Missing data on control variables was addressed using multiple imputation with chained equations in Stata 15’s mim suite (StataCorp15, College Station, TX). Twenty imputed datasets were created. Standard errors were calculated using Rubin’s rules (Rubin 1987). Table 1 shows descriptive statistics on the study variables, and Appendix A shows bivariate correlations between them.

Table 1.

Descriptive Statistics for Study Variables (N = 19,405 observations on 8,701 respondents)

	Mean/ Percent	SE	Min	Max
Latent class delinquency variety-depressive symptoms groups (9th grade)
1. Above-average delinquency variety	4.3%		0	1
2. Above-average on both (delinquency variety and depressive symptoms)	3.2%		0	1
3. Above-average depressive symptoms	10.5%		0	1
4. Average delinquency variety	17.0%		0	1
5. Low on both (delinquency and depression)	65.0%		0	1
Focal dependent variables (10th-12th grade)
Delinquency variety	1.180	0.015	0	9
Police contact	9.4%		0	1
Substance use	0.934	0.008	−0.069	3.475
Focal mediating variables (10th-12th grade)
Peer delinquency	1.173	0.010	0	9
Peer police contact	1.157	0.003	1	5
Peer substance use	0.934	0.006	−0.069	3.475
Control/background variables
Wave	−0.109	0.006	−1	1
Male	46.7%		0	1
White	86.0%		0	1
Black	2.6%		0	1
Hispanic	6.3%		0	1
Other non-white race	5.1%		0	1
Two biological parent family	63.2%		0	1
Eligible for free/reduced price lunch	18.3%		0	1
Poor family relations	−0.594	0.003	−1.830	0.914
Church attendance	2.280	0.009	1	4
Sensation seeking	2.286	0.007	1	5
School grades	4.074	0.006	1	5
School attachment	3.637	0.005	1	5
Police contact (9th grade)	6.5%		0	1
Substance use (9th grade)	0.586	0.007	−0.069	3.475
Peer delinquency (9th grade)	1.186	0.008	0	9
Peer police contact (9th grade)	1.125	0.002	1	5
Peer substance use (9th grade)	0.638	0.004	−0.069	3.475

Source: PROSPER Peers

Measures

Depressive symptoms in 9th grade.

The measure of depressive symptoms is the average of 5 items assessing frequent crying, deliberate self-harm, feelings of worthlessness, excessive guilt, and sadness over the past six months (each scored 0 to 2; α = .84). Higher scores indicate higher levels of depressive symptoms. This measure captures symptoms, not clinical diagnoses. The study results do not necessarily generalize to adolescents with diagnoses of depression.

Delinquency variety in 9th grade.

The measure of delinquency variety was created as the sum of 9 different dichotomized delinquency items adapted from the National Youth Survey (NYS; Elliott, Huizinga, and Menard 1989). Variety scale scaling was chosen because these scales are not dominated by high-frequency trivial acts (unlike frequency scales) and have high reliability and validity (Sweeten 2012). Items assessed how many times in the past 12 months respondents had purposely damaged property, avoided paying for things, stolen something from a store, stolen something worth less than $25, stolen something worth more than $25, broken into a building, thrown rocks or bottles at someone, beaten up or physically fought someone, and carried a hidden weapon (responses were originally coded 1 = never, 2 = once, 3 = twice, 4 = three or four times, and 5 = five or more times; α = .85). The items were dichotomized (0 = never, 1 = once or more) and summed them to create a count of the number of different delinquent acts respondents engaged in. Higher scores indicate higher delinquency variety.

Delinquency variety in 10th-12th grades.

The outcome measure of delinquency variety was created in the same way as the measure described above (α = .87).

Police contact in 10th-12th grades.

The outcome measure of police contact was a single item that asked respondents how many times in the past 12 months they had been picked up by the police for breaking a law (response categories ranged from 1 = never, 2 = once, 3 = twice, 4 = three or four times, and 5 = five or more times); the item was dichotomized (0 = no, 1 = yes).

Substance use in 10th-12th grades.

The measure of substance use comes from four items that assessed how often in the past month respondents had smoked any cigarettes, had beer, wine, or other hard liquor, got drunk from alcohol, and smoked marijuana (responses were coded 1 = not at all, 2 = one time, 3 = a few times, 4 = about once a week, 5 = more than once a week; α = .84). Item response theory (IRT) scaling (Osgood, McMorris, and Potenza 2002) was used to combine the items. Higher scores reflect higher levels of substance use.

Peer delinquency in 10th-12th grades.

The peer measures were assessed from 9th through 12th grade; the 10th-12th grade measures were examined as mediators. PROSPER used friendship nomination questions to assess peer ties. At each wave, respondents nominated up to seven friends in their same school and grade. The PROSPER staff matched over 83% of friendship nominations to students on the school rosters. Most (86%) of the unmatched nominations resembled no name on the class rosters and presumably were not respondents’ grademates. Since the entire grade participated in the study, these data allowed us to construct complete within-grade school friendship networks. The first step in the creation of the peer measures was the identification of students who named the respondent as a friend or were named as a friend by the respondent (i.e., undirected ties or the send-or-receive network). Respondents were then linked to their friends’ scores on items on the in-school survey. Peer delinquency is friends’ average score on the 9-item variety score of delinquency discussed above. On this and the other two peer measures, higher scores reflect higher levels of peer involvement in deviance.

Peer police contact in 10th-12th grades.

Peer police contact is friends’ average score on the ordinal item assessing police contact discussed above.

Peer substance use in 10th-12th grades.

Peer substance use is friends’ average score on the 4-item IRT scale of substance use discussed above.

Control variables.

The analyses accounted for several demographic and background characteristics that potentially could confound the results.

Wave.

The analyses controlled for wave and wave² to account for non-linear growth in the outcomes. This also included controls for respondents’ means over time on these variables to ensure that the time-varying versions capture the within-individual growth curve (Osgood 2010).

Male.

Male gender is included as a dichotomous control variable (0 = female, 1 = male).

Race.

The race/ethnicity control variables are represented as a set of dichotomous indicators for white, black, Hispanic, and other non-white race (for each, 0 = no, 1 = yes).

Two biological parent family.

This is a dichotomous control variable (0 = other family structure, 1 = living with two biological parents). This and the other control variables—with the exceptions of male gender, race, and baseline scores on the outcomes—are time-varying and assessed from 10th-12th grades.

Eligible for free or reduced price lunch.

This is a dichotomous control variable (0 = not eligible, 1 = eligible).

Poor family relations.

This control variable is the average of four standardized subscales that capture parent-child joint activities, affective quality, parental supervision, and family cohesion (α = .82). Example subscale items assessed the extent to which the respondent’s mom acted loving and affectionate toward him or her, the extent to which the respondent’s parents knew where he or she was during the day, and the past month frequency with which the respondent and a parent had done a fun activity that they both enjoyed. Higher scores reflect poorer quality family relations.

Church attendance.

Church attendance is a single item assessing how often respondents attended church or religious services in the past year (responses ranged from 1 = never to 4 = once a week or more).

Sensation seeking.

Sensation seeking is the mean of three items measuring respondents’ preference for risky and/or sensation seeking experiences (e.g., “I do what feels good regardless of the consequences”; responses ranged from 1 = never to 5 = always; α = .77).

School grades.

This variable is a single item tapping respondents’ grades in the past school year (responses ranged from 1 = mostly lower than Ds to 5 = mostly As).

School attachment.

The school attachment scale is the mean of 8 items on whether respondents held a positive attitude towards their school and teachers (e.g., “I like school a lot”; responses ranged from 1 = never true to 5 = always true; α = .77).

Adolescent deviance at 9th grade.

The analyses also included controls for respondents’ 9th grade levels of police contact, substance use, peer delinquency, peer police contact, and peer substance use when predicting those respective outcomes; these controls were identical in coding to the variables described above.

Analytical Strategy

The analytical strategy had three steps. First, the above measures and LCA were used to classify respondents into delinquency variety-depressive symptoms groups in 9th grade. The LCA was conducted with the gsem command available in Stata 15 to assess group membership. These models identified groups of respondents with similar scores on 9th grade depressive symptoms and delinquency variety. Linear and binomial models were used to account for the continuous and count distribution of the depressive symptoms and delinquency variety variables respectively. In exploratory analyses, one- through seven-group solutions were tested to determine the best-fitting model. Following the recommendations of others (e.g., Lanza and Rhoades 2013; Nagin 2005), the best-fitting model was identified through the use of the log-likelihood value (Log-L), Akaike’s information criterion (AIC), the Bayesian information criterion (BIC), and average posterior probabilities (AvePP), in conjunction with a consideration of parsimony, conceptual meaningfulness, and interpretability of the different classes.

Second, these groups were used as a categorical variable predicting peer delinquency, peer police contact, and peer substance use in 10–12th grade with linear random effects models. Taking peer delinquency as an example, the equation was as follows (where i indicates “this respondent,” t indicates “this wave,” and X and Z indicate vectors of time-varying and time-stable control variables respectively):

$$\mathit{\text{peerdelinquenc}}{y}_{\text{tij}}={\beta }_{\text{0ij}}+{\beta }_1\mathit{\text{latentgrou}}{p}_{\text{ij}}+{\beta }_2\mathit{\text{wav}}{e}_{\text{tij}}+{\beta }_3\mathit{\text{wav}}{e}_{\text{tij}}{}^2+\beta {\text{X}}_{\text{tij}}+\beta {\text{Z}}_{\text{ij}}+{\text{e}}_{\text{tij}}+r_{\text{ij}}+{\mu }_{\text{j}}$$

These models assessed whether the combination of delinquency variety and depressive symptoms predicted peer deviance differently than did either problem alone. Tests indicated that a polynomial for wave was not needed in the model predicting peer substance use, so the squared term was omitted there. In addition, tests revealed a significant latent group by wave interaction in the peer substance use model, so that interaction term was included in the final model.

Third, a series of random effects models was estimated predicting respondents’ delinquency variety, police contact, and substance use in 10–12th grade from the latent groups, focal mediators, and controls. This step involved two sub-steps. In the first sub-step, the models included just the latent groups and controls. The equations for this step resembled the equation presented above, except squared terms for wave were not needed in any of the three models. In the second sub-step, the focal mediators (peer deviance) were entered into the model. Taking respondent substance use as an example, the equation was as follows:

$$\mathit{\text{respondentsubstanceus}}{e}_{\text{tij}}={\beta }_{\text{0ij}}+{\beta }_1\mathit{\text{latentgrou}}{p}_{\text{ij}}+{\beta }_2\mathit{\text{peersubstanceus}}{e}_{\text{tij}}+{\beta }_3\mathit{\text{wav}}{e}_{\text{tij}}+\beta {\text{X}}_{\text{tij}}+\beta {\text{Z}}_{\text{ij}}+{\text{e}}_{\text{tij}}+r_{\text{ij}}+{\mu }_{\text{j}}$$

By providing estimates of β₁ both with and without the peer variables controlled, these sets of models allowed us to assess whether peer behavior mediated the effects of the latent groups. Given the limited and categorical nature of respondents’ delinquency variety and police contact, binomial regression and logistic regression were used when predicting those respective outcomes; respondents’ IRT substance use scores were treated as linear and continuous. Tests revealed a significant latent group by wave interaction in the delinquency model, so that interaction term was included in that final model.

The random effects models had three levels: waves (level 1), individuals (level 2), and school districts (level 3). The models corrected for this clustered data structure by including a variance component for each level of analysis. In addition, a variance component was included for wave (when needed) to account for residual dependence produced by serial autocorrelation.

Results

The first step of the analyses was a latent class analysis of delinquency variety and depressive symptoms at 9th grade. Although the Log-L, AIC, and BIC suggested that a six-group model best fit the data (see Appendix B for fit statistics), that solution resulted in empty classes and a poor AvePP. The next best-fitting model was the five-group model. The AvePP of assignment in that model was .89 across groups (ranging from .84 to .95), which is above the recommended minimum threshold of .70 (Nagin 2005). Furthermore, the five-group model resulted in conceptually meaningful groups that differed with respect to depressive symptoms and delinquency variety.

Labels for the groups were based on their mean depressive symptoms and delinquency variety relative to the sample means (.275 and 1.218 respectively; see Appendix C). Throughout this section, the word average is used to indicate that the group’s mean depressive symptoms or delinquency variety was within 1 SD of the mean; the phrase above-average indicates that the group’s mean was more than 1 SD from the mean; and the word low indicates that the group’s mean was near 0.

The above-average delinquency variety group (group 1), which made up approximately 4% of the sample, had delinquency variety scores well over 1 SD above the sample mean (M=7.066) and average depressive symptoms (M=.228). The above-average on both, or combined, group (group 2), which accounted for 3% of the sample, also had elevated levels of delinquency variety (M=5.650) but in addition had levels of depressive symptoms that were over 1 SD above the sample mean (M=1.439). The paper’s main focus is on the differences between delinquency variety alone versus in combination with depressive symptoms, so the main comparisons focus on these two groups. T-tests revealed that they had differing levels of both delinquency variety and depressive symptoms. The difference in delinquency variety, though significant, was modest (0.71 SD), whereas the difference in depressive symptoms was larger (2.86 SD). Figure 1 illustrates the mean delinquency variety and depressive symptoms sores of these and the other groups.

Fig. 1.

Mean delinquency and depression scores by group membership

NOTE: avg. = average, delinq. = delinquency, depress. = depression.

The other four groups had various levels of depressive symptoms and delinquency variety. The above-average depressive symptoms group (group 3; 11%) had above-average depressive symptoms (M=1.084) but average delinquency variety (M=.751). The average delinquency variety group (group 4; 17%) had average delinquency variety scores (M=3.090) and low depressive symptoms (M=.190). Finally, low on both (group 5) was the largest group, comprising 65% of the sample; this group had few depressive symptoms (M=.112) and little delinquency (M=.198) in 9th grade.

In addition to showing the levels of 9th grade depressive symptoms and delinquency variety among the latent classes, Appendix C provides background information about the classes. The two main groups of interest differed from each other on some characteristics measured at 9th grade. Specifically, members of the above-average on both group (group 2) were less likely to be male, were more likely to be of other non-White race, had lower sensation-seeking, and had lower baseline peer delinquency and peer police contact than the above-average delinquency variety group (group 1).

Associations of Delinquency Variety-Depressive Symptoms Categories with Peer Deviance

Table 2 shows the associations of 9th grade delinquency variety-depressive symptoms category membership with peer delinquency, peer police contact, and peer substance use across 10th-12th grades (coefficients for control variables are shown in Appendix D). Relative to respondents who only showed above-average delinquency variety (group 1), those who showed both above-average delinquency variety and above-average depressive symptoms—the combined group (group 2)—went on to have friends who had fewer police contacts (b = −0.079, p < .001), and engaged in less substance use (b when wave equals 0 [i.e., 11th grade] = −0.161, p < .01), although their friends were no more or less delinquent. A Wald test confirmed that in the model predicting peer substance use, the above-average on both coefficient and the interaction term between that coefficient and wave (b = −0.061, p > .05) were jointly statistically significant (F = 4.80, p < .001). Post hoc calculations revealed that groups 1 and 2 differed by 21% SD on later peer police contact and 12%, 20%, and 27% SD on peer substance use at 10th, 11th, and 12th grades respectively. Thus co-occurring delinquency and depression were associated with less deviant peer affiliation than elevated delinquency only.

Table 2.

Linear Random Effects Coefficients Predicting 10th-12th Grade Peer Delinquency, Police Contact, and Substance Use from Latent Classes of 9th Grade Delinquency Variety and Depressive Symptoms (N = 19,405 observations on 8,701 respondents)

	Peer Delinquency				Peer Police Contact			Peer Substance Use
Predictor	b		(se)		b	(se)		b		(se)
1. Above-average delinquency variety (ref.)	--				--			--
2. Above-average on both	−0.111		(0.081)		−0.079	(0.023)	***	−0.161		(0.050)	**
3. Above-average depressive symptoms	−0.257		(0.063)	***	−0.084	(0.018)	***	−0.143		(0.039)	***
4. Average delinquency variety	−0.128		(0.058)	*	−0.057	(0.017)	***	−0.021	b	(0.036)
5. Low on both	−0.277	a	(0.055)	***	−0.081	(0.016)	***	−0.101		(0.034)	**
Above-average on both * wave	--				--			−0.061		(0.046)
Above-avg. depressive symptoms * wave	--				--			−0.096		(0.034)	**
Average delinquency variety * wave	--				--			−0.006		(0.032)
Low on both * wave	--				--			−0.046		(0.030)

Source: PROSPER Peers

^ap<.01 for difference from above-average on both group

^bp<.001 for difference from above-average on both group

^†p<.10;

^*p<.05;

^**p<.01;

^***p<.001

As table 2 also shows, respondents who had above-average depressive symptoms only, who showed average levels of delinquency variety, or who showed low levels of both delinquency variety and depressive symptoms—that is, those in the remaining groups—also had less deviant friends than respondents who showed above-average delinquency variety scores. Relative to the combined group (group 2), the average delinquency group showed more peer substance use, and the low on both group showed less peer delinquency.

Associations of Delinquency Variety-Depressive Symptoms Categories with Respondents’ Own Deviance

We next examined whether delinquency variety-depressive symptoms category membership at 9th grade predicted involvement in deviance from 10th-12th grades. Table 3 shows these estimates (control coefficients are shown in Appendix E). Relative to the above-average delinquency variety category (group 1), respondents in the combined delinquency variety and depressive symptoms group (group 2) had lower delinquency variety (model 1; b at 11th grade = −0.298, p < .05), but a significant interaction between this group and wave (b = 0.307, p < .001) revealed that this was only true at 10th and 11th grades. The combined group also had lower substance use (model 5; b = −0.173, p < .05), but they did not have lower odds of police contact (model 3; b = −0.093, p > .05). The coefficients predicting delinquency (at 11th grade) and substance use from membership in group 2 translate to a 15% SD difference in each of those outcomes.

Table 3.

Random Effects Coefficients Predicting 10th-12th Grade Delinquency, Police Contact, and Substance Use from Latent Classes of 9th Grade Delinquency Variety and Depressive Symptoms and Peer Delinquency (N = 19,405 observations on 8,701 respondents)

	Delinquency						Police Contact						Substance Use
	Model 1			Model 2			Model 3			Model 4			Model 5			Model 6
Predictor	b (se)			b (se)			b (se)			b (se)			b (se)			b (se)
Latent class delinquency variety-depressive symptoms groups
1. Above-average delinquency variety (ref.)	--			--			--			--			--			--
2. Above-average on both	−0.298 (0.137)		**	−0.277 (0.135)		*	−0.093 (0.209)			−0.042 (0.207)			−0.173 (0.060)		**	−0.127 (0.060)
3. Above-average depressive symptoms	−1.819 (0.111)	c	***	−1.762 (0.110)	c	***	−1.172 (0.189)	c	***	−1.118 (0.187)	c	***	−0.121 (0.050)		*	−0.120 (0.047)
4. Average delinquency variety	−0.934 (0.099)	c	***	−0.907 (0.098)	c	***	−0.386 (0.154)		*	−0.356 (0.153)	a	*	0.031 (0.046)	c		0.003 (0.043)	b
5. Low on both	−2.524 (0.094)	c	***	−2.461 (0.094)	c	***	−1.347 (0.155)	c	***	−1.299 (0.154)	c	***	−0.105 (0.045)		*	−0.122 (0.042)
Above-average on both * wave	0.307 (0.069)		***	0.329 (0.069)		***	--			--			--			--
Above-avg. depressive symptoms * wave	0.396 (0.057)		***	0.417 (0.057)		***	--			--			--			--
Average delinquency variety * wave	0.281 (0.048)		***	0.291 (0.048)		***	--			--			--			--
Low on both * wave	0.440 (0.047)		***	0.451 (0.047)		***	--			--			--			--
Focal mediating variables
Peer delinquency				0.105 (0.009)		***
Peer police contact										0.593 (0.069)		***
Peer substance use																0.295 (0.009)		***

NOTE: Binomial, logistic, and linear coefficients shown for delinquency, police contact, and substance use models respectively.

Source: PROSPER Peers

^ap<.10 for difference from above-average on both group

^bp<.01 for difference from above-average on both group

^cp<.001 for difference from above-average on both group

^†p<.10;

^*p<.05;

^**p<.01;

^***p<.001

Next, models 2, 4, and 6 repeated the same analyses, adding peer deviance as a mediator. An examination of the coefficients from models 1 and 2 indicated that the addition of peer delinquency to the model had little effect on the difference in delinquency between the above-average delinquency variety group and the combined group. For example, the coefficient at 11th grade (b = −0.298) was reduced by only 7% (to b = −0.277) when peer delinquency was added, and the significance level was unchanged. In addition, as noted above, model 3 indicated that there was no evidence of any difference to be mediated between the above-average delinquency group (group 1) and the combined group (group 2) on police contact. There thus was little evidence of mediation for these two outcomes.

There was evidence of partial mediation for the substance use outcome. A comparison of models 5 and 6 showed that peer substance use explained 27% of the difference in substance use between the combined group and the above-average delinquency variety group. These results suggest that deviant peer affiliation partly explained why adolescents with only elevated delinquency engaged in more future substance use than adolescents who also had elevated depressive symptoms.

Finally, table 3 also shows that respondents with above-average depressive symptoms and delinquency variety (group 2) had significantly higher levels of future delinquency than all three of the other groups, and had higher levels of future police contact than respondents who were low on both (group 5) or who had elevated depression only (group 3), although they had lower levels of future substance use than respondents who showed average levels of delinquency variety (group 4). These findings indicate that the combined group is still at risk for negative future outcomes.

Alternate Specifications

Several alternative models were estimated as well. These models showed that the results were generally robust to different codings of the focal variables and different types of regression, and that overall the findings did not differ by gender. First, the analyses were repeated using peer measures that were based only on reciprocated friendship ties (versus all ties as in the main models). The logic was that reciprocated ties represent stronger, and potentially more influential, friendships. The only substantive difference between these models and the main models was that in the reciprocated tie models, the combined group and the above-average delinquency group did not differ on peer police contact. Still, these groups did differ on peer delinquency and substance use and on respondents’ own delinquency and substance use, and peer substance use explained 24% of the difference in respondents’ own substance use.

Second, the fact that the mediators and outcomes were measured during the same waves raises the possibility of reverse causality between them. As a robustness check, the model in which there was evidence of mediation—the substance use model—was re-estimated using 9th grade latent groups, 10th grade peer substance use, and 11th-12th grade respondent substance use as measures. The results revealed that the above-average on both group showed lower substance use than the above-average delinquency group at 11th-12th grades, and that 10th grade peer substance use reduced that difference by approximately 12%. The results thus were less pronounced but similar under this more stringent timing of measurement.

Third, the main analyses used a dichotomous measure of respondents’ own police contact as an outcome. As a robustness check, this model was repeated using an ordinal measure of respondents’ own police contact (with response categories ranging from 1 = never, 2 = once, 3 = twice, 4 = three or four times, and 5 = five or more times) and ordinal random effects regression. The results of that model were substantively similar to dichotomous measure in that the combined group still did not differ significantly from the above-average delinquency group.

Fourth, the delinquency analyses were repeated using negative binomial rather than binomial regression. In these models, although peer delinquency was a significant predictor of respondents’ delinquency, membership in the combined group was not. These results thus also supported the conclusion that co-occurring depression did not influence later delinquency through delinquent peer affiliations.

Finally, to determine whether gender moderated the results, the models were re-estimated separately for males and females and the equivalence of coefficients across models was tested. Of the 32 resultant comparisons, two differences were statistically significant. These differences suggested that the finding that the combined group’s friends had lower rates of police contact may apply only to males, and that wave and membership in the combined group may interact to predict delinquency only among males. Because of the large number of significance tests run, readers should interpret these few significant interactions with caution.

Discussion

Although many delinquent youth also have symptoms of depression, past research was mixed on whether those symptoms have any impact on the future delinquency and substance use of those youth. Past work also had not yet fully explored the role of peers in this potential impact, even though deviant peer affiliations play a major role in the adolescent escalation of delinquency and theoretically should be influenced by depression (Hussong et al. 2011). This study’s objective was to test whether co-occurring depressive symptoms increased or decreased future deviance among already delinquent youth, and whether it did so by increasing or decreasing affiliations with peers who were involved in those same deviant behaviors. The results showed that co-occurring depressive symptoms protected delinquent youth against exposure to peers who got into trouble with the police or who used drugs and alcohol. They also protected them against engaging in future delinquency and substance use themselves. However, peer deviance did not account for the association of co-occurring depressive symptoms with future delinquency, and it accounted for only a modest portion of the association of depressive symptoms with future substance use. In addition, co-occurring depressive symptoms did not protect delinquent youth against future contact with the police, even though they protected them against having friends who had had police contact. Thus, although the study provides support for a protective role of co-occurring depressive symptoms, it provides only weak support for a full pathway from those symptoms to fewer deviant friends to reductions in future deviance.

This study’s results are in line with those of the two most similar past studies on the topic (Mason et al. 2008; Scalco et al. 2014), despite its different methodology. Note that those studies examined only substance use, and found evidence that peers partially mediated the pathway from depressive symptoms to substance use. This study found the same, but it also found that peer deviance did not mediate the pathway from depressive symptoms to other forms of deviance. It is possible that this reflects differences in the types of peers who engage in, and thus model, delinquency versus substance use. For example, adolescent friendship groups with high levels of alcohol use also have high network status and internal cohesion, whereas adolescent friendship groups with high levels of delinquency have lower status and cohesion (Kreager, Rulison, and Moody 2011). If co-occurring depressive symptoms lead to withdrawal specifically from high status peer groups, this could explain why they have different impacts on different forms of deviance. Future research should examine in more detail the mechanisms by which symptoms influence peer affiliations, and the implications of those mechanisms for specific behaviors.

The study’s findings are also in line with those of other studies that have shown a beneficial effect of co-occurring problems on adolescent substance use (Colder et al. 2018; Stone, Vander Stoep, and McCauley 2016), and with a recent meta-analysis, which suggested that comorbid internalizing and externalizing disorders are a somewhat weaker predictor of recidivism than are “pure” externalizing disorders (Wibbelink, Hoeve, Stams, and Oort 2017). However, they may not appear to mirror another study’s finding that depressed mood alone does not predict serious delinquency among mid- to late adolescents (Stouthamer-Loeber, Loeber, Farrington, and Wikström 2002), or a third study’s finding that internalizing problems reduce arrest (Hirschfield, Maschi, White, Traub, and Loeber 2006). The difference could stem from the studies’ focuses. The current study sought to compare the outcomes of already delinquent youth who did and did not also have depressive symptoms; this was the key contrast that the latent class analysis helped isolate. Still, relevant to the other studies’ findings, in the current study the co-occurring problems group showed more future deviance than did any group without elevated delinquency, including the group that experienced only depressive symptoms. This mirrors past results on the overall negative outcomes of youth who experience comorbid problems (Sourander et al. 2007).

This study provides conflicting evidence for the theoretical pathways linking co-occurring delinquency and depressive symptoms and criminogenic outcomes and processes (Damphousse and Kaplan 1998; Hussong et al. 2011). Most notably, they cannot tell us why reduced deviant peer affiliations did not always translate to less delinquency and substance use among adolescents who experienced both problems. This is, after all, one of the main theoretical mechanisms that has been proposed. One possibility is that the co-occurring group is misperceiving their friends’ levels of deviance and being influenced by those misperceptions rather than by their friends’ actual behavior. Indeed, adolescents with more depressive symptoms are more likely to perceive that their peers use drugs and alcohol (Siennick, Widdowson, Woessner, Feinberg, and Spoth 2017). Future research should consider whether and how perceived norms and other risk factors for persistent delinquency differ for adolescents with different constellations of problems.

The current findings raise the question of why else, if not because of peer deviance, co-occurring depressive symptoms would reduce future deviance among already delinquent youth. It is possible that the symptoms impact other aspects of peer relations, such as popularity, that themselves are associated with delinquency. It also is possible that depressive symptoms impact one of the several other major factors thought to explain developmental change in delinquency, such as victimization, impulse control, emotion regulation, family ties, or school involvement (Sweeten, Piquero, and Steinberg 2013). Future research should continue to examine potential mediators of the longitudinal pathway from co-occurring problems to persistent adolescent delinquency.

This study’s findings also raise the question of why, if depressive symptoms improve deviant outcomes among delinquent youth, depression and delinquency are so likely to positively co-occur in the first place. There are several potential reasons. As noted above, the failure model still could apply, if delinquency leads to social rejection which in turn leads to depression. Shared risk, that is, common root causes between the two problems, is another likely possibility. This study’s results yield less suggestive evidence for the third broad possibility, the acting-out model, which posits that depression creates internal pressure to engage in deviance. Still, either of the former possibilities could explain why so many studies have found a cross-sectional association between deviant behavior and depressive symptoms. Notably, this study examined only the contribution of co-occurring depression to later deviance, not the contribution of co-occurring delinquency to later depression. Scholars should continue to examine the concurrent association between delinquency and depressive symptoms, but they also should examine whether youth with both problems go on to experience more or less future depression.

Although this study suggests that depressive symptoms can protect delinquent youth against continued delinquency, this does not mean that those symptoms are “good for” youth. Other studies have linked the combination of delinquency and depressive symptoms to negative outcomes such as suicide attempts and self-harm behavior (Kang et al. 2015; Penn, Esposito, Schaeffer, Fritz, and Spirito 2003). Thus, there is a clear need for interventions among this group of adolescents. Yet, the current study indicates that this may be complicated by the potential protective effect of depressive symptoms on affiliation with deviant peers. That is, treating depressive symptoms (alone) among already delinquent adolescents might unintentionally increase some negative risk exposures like affiliation with peers involved in delinquency, the justice system, and substance use. Interventions aimed at treating this group may need to consider and guard against such unintended harmful consequences.

This study had several methodological strengths, including the examination of multiple deviant outcomes, the use of a measurement strategy that isolated the comparison between youth with both problems and youth with only elevated delinquency, and the use of peers’ own reports of their deviant behaviors. Still, the study was not without limitations. First, the two focal latent groups differed not only on baseline depressive symptoms, but also on baseline delinquency. It is possible that the protective “effect” of co-occurrence on peer deviance was due to the modestly lower level of delinquency in the co-occurring group. Future studies could usefully examine outcomes among similar groups with more comparable levels of baseline delinquency. Second, although this study addressed temporal ordering between the co-occurrence and later outcomes, and the findings were confirmed using more stringent timings of measurement of the mediators and outcomes, the findings might differ depending on the duration that adolescents experienced both delinquency and depressive symptoms (McCarty et al. 2013). Future research should examine whether the consequences of co-occurring delinquency and depressive symptoms differ depending on how long adolescents experienced each. Third, in the latent class analysis, the residual variances were constrained to be equal across classes. Attempts to relax this assumption led to problems with model convergence. It is possible that the results would have differed if the variances were left unconstrained. Fourth, the data did not include measures of other potentially important confounds, such as temperament, maternal depression, and family history of substance abuse problems (Fanti and Heinrich 2010; Pardini, White, and Stouthamer-Loeber 2007). Other researchers should examine whether the findings hold when a more comprehensive set of background factors is controlled. Fifth, the study’s statistical models were fairly complex, and significance tests for mediation under our modeling strategy have not yet been developed. Thus the degree of mediation found should be interpreted with caution. Finally, the data were collected in largely rural and predominately White areas with large proportions of low-income families. Thus, the nature of the sample could limit the generalizability of the findings. Future research should replicate these findings to determine whether similar results emerge in different populations and contexts.

Conclusion

Adolescence is a high-risk time for the emergence of depressive symptoms and for the escalation of delinquency and substance use. Although past research had considered why depression and delinquency are so likely to co-occur among adolescents, it had not yet fully examined whether co-occurring depressive symptoms influence persistence in deviance among delinquent youth. This study tested whether those symptoms increase or decrease future deviance by altering friendships with deviant peers. The results showed that co-occurring depressive symptoms protected delinquent youth against deviant peer affiliations, and also lowered the youth’s own levels of delinquency and substance use. However, deviant peer affiliations were a weak explanation for why youth with co-occurring problems reduced their own involvement in deviance. The different outcomes of adolescents experiencing co-occurring delinquency and depressive symptoms relative to adolescents experiencing “pure” delinquency and “pure” depressive symptoms suggest the potential need for customized interventions for this group. To better inform these interventions, research should continue to examine the outcomes of these youth, and what, if not deviant peers, explains those outcomes.

Biography

Sonja E. Siennick is a professor at Florida State University’s College of Criminology and Criminal Justice. She studies criminal offending and mental health problems in the contexts of the life course as well as kinship and friendship relations.

Alex O. Widdowson is an assistant professor of criminal justice at the University of Louisville. He studies the development of crime over the life course, the consequences of criminal behavior and criminal justice sanctioning, peer delinquency, and prisoner reentry.

Mark E. Feinberg is a research professor in the Prevention Research Center at The Pennsylvania State University. He has made significant contributions to theory and research on coparenting, sibling relationships, dynamic modeling of family interaction, the epidemiology of adolescent behavior problems, family violence, and the functioning and efficacy of community-level public health initiatives. He also developed the Family Foundations.

Appendix A.

Bivariate Correlations between Study Variables

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23
1.Delinquency	1.00
2.Police contact	0.49	1.00
3.Substance use	0.42	0.30	1.00
4.Peer delinquency	0.23	0.15	0.22	1.00
5.Peer police cont.	0.14	0.15	0.17	0.56	1.00
6.Peer subst. use	0.19	0.18	0.44	0.46	0.37	1.00
7.Wave	−0.04	−0.01	0.10	−0.06	−0.02	0.15	1.00
8.Male	0.15	0.08	0.02	0.17	0.11	0.03	0.00	1.00
9.Black	0.00	0.00	−0.02	0.03	0.02	0.00	−0.01	0.00	1.00
10.Hispanic	0.00	0.00	−0.03	−0.02	−0.01	−0.04	−0.01	−0.03	−0.04	1.00
11.Other race	0.06	0.03	0.01	0.03	0.02	0.01	−0.01	0.03	−0.04	−0.06	1.00
12.Two bio. parents	−0.09	−0.07	−0.11	−0.08	−0.05	−0.08	0.01	0.04	−0.05	−0.01	−0.04	1.00
13.Free lunch	0.05	0.03	0.00	0.05	0.03	−0.01	−0.05	−0.02	0.08	0.22	0.01	−0.22	1.00
14.Family relations	0.28	0.15	0.23	0.10	0.06	0.09	−0.01	0.04	−0.02	−0.02	0.03	−0.05	0.01	1.00
15.Church attend.	−0.12	−0.07	−0.18	−0.12	−0.07	−0.15	−0.06	−0.03	0.02	0.02	−0.02	0.20	−0.08	−0.18	1.00
16.Sens. seeking	0.40	0.23	0.34	0.18	0.12	0.17	−0.03	0.18	−0.01	−0.03	0.04	−0.08	0.01	0.26	−0.14	1.00
17.School grades	−0.21	−0.17	−0.22	−0.18	−0.14	−0.16	0.09	−0.15	−0.04	−0.08	−0.03	0.21	−0.17	−0.16	0.19	−0.21	1.00
18.School attach.	−0.33	−0.21	−0.29	−0.19	−0.13	−0.16	0.03	−0.17	0.00	0.01	−0.03	0.12	−0.04	−0.40	0.21	−0.34	0.45	1.00
19.Police contact	0.22	0.21	0.17	0.10	0.08	0.10	−0.02	0.08	0.01	0.01	0.06	−0.06	0.03	0.09	−0.06	0.12	−0.12	−0.14	1.00
20.Substance use	0.31	0.24	0.50	0.17	0.13	0.32	−0.04	−0.02	−0.04	−0.01	0.01	−0.10	0.01	0.16	−0.15	0.23	−0.19	−0.21	0.31	1.00
21.Peer delinquency	0.21	0.14	0.19	0.29	0.16	0.22	−0.03	0.19	0.09	0.03	0.04	−0.10	0.08	0.10	−0.15	0.17	−0.21	−0.19	0.13	0.21	1.00
22.Peer police cont.	0.12	0.12	0.15	0.15	0.14	0.17	−0.03	0.11	0.04	0.00	0.02	−0.06	0.05	0.06	−0.08	0.09	−0.17	−0.12	0.14	0.17	0.53	1.00
23.Peer subst. use	0.19	0.16	0.34	0.21	0.15	0.40	−0.04	−0.03	−0.01	−0.04	0.00	−0.10	0.02	0.09	−0.16	0.15	−0.17	−0.15	0.14	0.41	0.51	0.40	1.00

NOTE: Variables 1–6 measured at 10th-12th grades; variables 19–23 measured at 9th grade.

Source: PROSPER Peers

Appendix B.

Goodness of Fit Statistics for Latent Class Models

Number of classes	Log-Likelihood	AIC	BIC	df	AvePP
1	−26764.75	53535.51	53556.99	3	--
2	−19924.74	39861.47	39904.44	6	0.965
3	−19307.86	38633.72	38698.17	9	0.910
4	−18097.31	36218.62	36304.56	12	0.896
5	−17198.84	34427.67	34535.10	15	0.891
6	−17198.92	34427.83	34535.26	15	0.743
7	−16345.69	32735.38	32892.94	22	0.861

Source: PROSPER Peers

Appendix C.

Means/Percentages on Depressive Symptoms, Delinquency Variety and Background Variables at 9th Grade by Latent Class Membership

	Mean	SD	Above-average delinquency variety	Above-average on both	Above-average depressive symptoms	Average delinquency variety	Low on both	Post-Hoc Tests
Group	All		1	2	3	4	5
Latent class delinquency-depression groups
Percent in this class			4.3%	3.2%	10.5%	17.0%	65.0%
9th grade depressive symptoms	0.275	0.423	0.228	1.439	1.084	0.190	0.112	a
9th grade delinquency variety	1.218	2.001	7.066	5.650	0.751	3.090	0.198	a
Background variables
Male			77.8%	32.9%	18.0%	59.5%	48.5%	a
Black			4.8%	2.2%	2.6%	4.1%	2.6%
Hispanic			9.0%	7.6%	7.4%	7.5%	6.7%
Other non-white race			7.4%	13.0%	5.9%	6.3%	4.9%	a
Two biological parent family			50.7%	46.9%	52.1%	53.5%	66.5%
Free/reduced price lunch			27.3%	23.5%	25.8%	23.6%	17.9%
Poor family relations			−0.347	−0.362	−0.475	−0.466	−0.665
Church attendance			1.984	2.054	2.183	2.064	2.448
Sensation seeking			3.095	2.953	2.403	2.684	2.138	a
School grades			3.499	3.590	3.871	3.781	4.098
School attachment			3.109	3.066	3.460	3.396	3.731
Police contact			43.8%	41.8%	4.5%	14.9%	2.4%
Substance use			1.760	1.766	0.746	1.082	0.376
Peer delinquency			2.114	1.757	1.181	1.688	1.089	a
Peer police contact			1.307	1.223	1.128	1.198	1.112	a
Peer substance use			0.962	0.938	0.699	0.848	0.587
N	8,701		377	277	911	1,476	5,660

Source: PROSPER Peers

^aSignificant difference (p<.05) between above-average delinquency variety (group 1) and above-average on both (group 2)

Appendix D.

Coefficients for Control Variables from Table 2

	Peer Delinquency			Peer Police Contact			Peer Substance Use
Predictor	b	(se)		b	(se)		b	(se)
Wave	−0.058	(0.010)	***	0.005	(0.003)		0.229	(0.029)	***
Wave2	−0.035	(0.017)	*	−0.009	(0.005)	†
Mean wave	−0.271	(0.044)	***	−0.081	(0.013)	***	−0.148	(0.026)	***
Mean wave2	−0.037	(0.100)		0.012	(0.029)
Male	0.264	(0.022)	***	0.052	(0.006)	***	−0.003	(0.013)
Black	−0.072	(0.065)		0.018	(0.019)		0.001	(0.040)
Hispanic	−0.098	(0.045)	*	−0.021	(0.013)	†	−0.027	(0.027)
Other non-white race	0.036	(0.047)		−0.002	(0.014)		0.009	(0.028)
Two biological parent family	−0.058	(0.022)	**	−0.006	(0.006)		−0.044	(0.013)	***
Free/reduced price lunch	0.048	(0.027)	†	0.008	(0.008)		−0.052	(0.015)	***
Poor family relations	0.039	(0.023)	†	−0.002	(0.007)		0.000	(0.014)
Church attendance	−0.042	(0.008)	***	−0.009	(0.002)	***	−0.022	(0.005)	***
Sensation seeking	0.083	(0.010)	***	0.018	(0.003)	***	0.056	(0.006)	***
School grades	−0.094	(0.014)	***	−0.029	(0.004)	***	−0.057	(0.008)	***
School attachment	−0.093	(0.017)	***	−0.019	(0.005)	***	−0.046	(0.010)	***
Outcome measure at 9th grade	0.221	(0.010)	***	0.123	(0.011)	***	0.468	(0.011)	***
Intercept	1.742	(0.121)	***	1.236	(0.038)	***	1.108	(0.061)	***

Source: PROSPER Peers

^†p<.10;

^*p<.05;

^**p<.01;

^***p<.001

Appendix E.

Coefficients for Control Variables from Table 3

	Delinquency				Police Contact				Substance Use
	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
Predictor	b (se)		b (se)		b (se)		b (se)		b (se)		b (se)
Wave	−0.029 (0.012)	*	−0.023 (0.012)	†	0.177 (0.041)	***	0.173 (0.041)	***	0.196 (0.007)	***	0.141 (0.007)	***
Mean wave	−0.208 (0.074)	**	−0.168 (0.074)	*	−0.608 (0.141)	***	−0.549 (0.140)	***	−0.192 (0.033)	***	−0.139 (0.032)	***
Male	0.438 (0.044)	***	0.398 (0.044)	***	0.235 (0.080)	**	0.190 (0.079)	*	−0.068 (0.017)	***	−0.067 (0.016)	***
Black	0.019 (0.124)		0.000 (0.123)		−0.220 (0.239)		−0.253 (0.238)		−0.027 (0.051)		−0.035 (0.048)
Hispanic	0.090 (0.084)		0.104 (0.083)		0.137 (0.156)		0.155 (0.154)		−0.058 (0.035)	†	−0.046 (0.033)
Other non-white race	0.334 (0.088)	***	0.324 (0.087)	***	0.073 (0.157)		0.074 (0.155)		−0.019 (0.037)		−0.020 (0.035)
Two biological parent family	−0.106 (0.039)	**	−0.103 (0.038)	**	−0.130 (0.077)	†	−0.126 (0.077)	†	−0.042 (0.017)	*	−0.027 (0.016)	†
Free/reduced price lunch	0.025 (0.039)		0.020 (0.038)		−0.008 (0.092)		−0.012 (0.092)		−0.077 (0.020)	***	−0.064 (0.019)	***
Poor family relations	0.678 (0.034)	***	0.671 (0.034)	***	0.440 (0.081)	***	0.441 (0.080)	***	0.165 (0.017)	***	0.166 (0.017)	***
Church attendance	0.005 (0.014)		0.011 (0.014)		−0.001 (0.031)		0.007 (0.031)		−0.039 (0.006)	***	−0.030 (0.006)	***
Sensation seeking	0.488 (0.014)	***	0.483 (0.014)	***	0.518 (0.035)	***	0.509 (0.035)	***	0.189 (0.008)	***	0.176 (0.007)	***
School grades	−0.085 (0.019)	***	−0.078 (0.019)	***	−0.321 (0.044)	***	−0.300 (0.044)	***	−0.078 (0.010)	***	−0.059 (0.009)	***
School attachment	−0.438 (0.024)	***	−0.422 (0.024)	***	−0.456 (0.060)	***	−0.436 (0.059)	***	−0.138 (0.013)	***	−0.123 (0.012)	***
Outcome measure at 9th grade					1.242 (0.120)	***	1.199 (0.119)	***	0.487 (0.010)	***	0.419 (0.009)	***
Intercept	0.072 (0.142)		−0.165 (0.143)		−0.527 (0.300)	†	−1.378 (0.313)	***	1.402 (0.075)	***	1.038 (0.071)	***

NOTE: Binomial, logistic, and linear coefficients shown for delinquency, police contact, and substance use models respectively.

Source: PROSPER Peers

^†p<.10;

^*p<.05;

^**p<.01;

^***p<.001