Individual,Social,and Neighborhood Influences on the Launch of Adolescent Antisocial Behavior

Abstract

This study attempts to develop further insight into the initiation and continuance of antisocial behavior during adolescence using a “launch” perspective in conjunction with several risk/protective factor measures from prior research on substance use and delinquency. Data from the Project on Human Development in Chicago Neighborhoods (PHDCN) were analyzed using multilevel latent growth curve models across three waves for three age cohorts (n = 2,130 youth; 78 neighborhoods). After considering typical trends and their variation, the relationships between covariates and different characteristics of the growth curves were analyzed. Implications of the findings for understanding developmental trends and prevention strategy are considered.

Keywords

developmental trends antisocial behavior launch model prevention

Adolescent risk behavior can have serious consequences for youths and can also impose costs to society more generally (Albert & Steinberg, 2011; Cohen & Piquero, 2009; Farrington & Welsh, 2007). Recent national trends in adolescent substance use suggest declines (Johnston, O'Malley, Bachman, & Schulenberg, 2011), but nearly 40% of 12th-grade students surveyed for the Monitoring the Future Study still report using some illicit drug. Similarly, national estimates show a reduction in delinquency cases from 1995 to 2007 (Knoll & Sickmund, 2010), but the prevalence of delinquency remains high and data show a flat trend more recently. These figures illustrate the continued importance of understanding the etiology of adolescent antisocial behavior—in particular, its initiation and continuation.

Developmental and Public Health Perspectives on Antisocial Behaviors

Although antisocial behavior can be studied from a variety of perspectives, a developmental, life-course framework have been utilized with increasing regularity in recent years (Farrington, 2005, 2006; Hser, Longshore, & Anglin, 2007). This approach draws on the study of individuals over time and sets the stage for the development of interventions informed by an understanding of the onset and continuance of antisocial behavior (Loeber & Farrington, 2000). In concert with this approach, some have begun to take a public health view of substance use and delinquency, with an emphasis on interventions that aim to direct youth away from antisocial behavioral trajectories (Welsh, 2005). This, however, requires a thorough understanding of the etiology of antisocial behavior at its initial point and across its developmental course (Jenson & Fraser, 2010). Recent analysis emphasizes the importance of pursuing prevention efforts at relatively early developmental stages to stem the monetary and social costs associated with long-term patterns of antisocial behavior (Cohen & Piquero, 2009). At the same time, a growing body of literature has identified effective primary and secondary prevention programs for individuals, families, and communities (Catalano, 2007).

The “Launch” of Adolescent Antisocial Behavior

This research has led to the compilation of comprehensive lists of individual and social risk factors found to increase the likelihood of antisocial behavior (e.g., Hawkins et al., 1998). Multiple frameworks can be used to understand the process by which this initial risk might affect long-term behavioral trajectories. These perspectives sometimes differ in their views of the nature of early risk factors and the extent and manner in which they affect later behavior. The “launch” model proposes that risk factors, in addition to their contemporaneous impact, have effects on later trends in antisocial behavior as well (Hussong, Curran, Moffitt, & Caspi, 2008; Kinderman & Skinner, 1992). So, a given influence (e.g., family disadvantage) is expected to impact a youth’s immediate behavior and also have an effect on their later developmental trend.

Hussong and colleagues (2008) used this perspective to study substance use and continuity in criminal activity in late adolescence/early adulthood. They found that substance use had a positive association with the initial level of offending and a negative effect on its slope, suggesting steeper declines in offending among those who had higher initial levels of alcohol and marijuana abuse symptoms. Those individuals still had comparatively high offending levels, however. An assessment of this explanatory model at earlier developmental periods is essential as it might offer insight into whether there are behaviors that are affected by risk factors both concurrently and prospectively, which is the notion underlying the launch model (Kinderman & Skinner, 1992), or, alternatively, whether later behavioral trends are relatively impervious to early influences.

Individual and Social Influences on the Development of Antisocial Behavior

Using this perspective as a jumping off point, an important next layer of investigation is to identify the factors that play a part in the developmental processes highlighted in the launch model. The previous literature suggests a number of possibilities that might be investigated in such a model. Specifically, researchers have considered several domains in assessing the etiology of problem behaviors in adolescence moving outward from the individual to immediate and more distal social influences. First, research and theory on between-individual differences that are influential in adolescent deviance often points to a behavioral “propensity” or “potential.” These factors, which include things like self-control and emotionality (Gottfredson & Hirschi, 1990; Lahey & Waldman, 2005), are believed to vary across individuals and usually comprise aspects of a youth’s personality and temperament that condition their actions in response to environmental stimuli.

While it is clear that individual constitutional factors affect antisocial behavior in adolescence, identification of such influences is unlikely to fully explain the emergence of antisocial behavior (Sampson & Laub, 1993); this requires the inclusion of social influences in developmental models. Among these influences, the immediate social context, reflected in family and peers, is prominent in theory and empirical research on adolescent antisocial behavior. Simons and colleagues (2004) summed up parental influence in terms of dimensions of demandingness and responsiveness, noting that parenting for positive developmental outcomes generally requires some of each. Loeber and Stouthamer-Loeber (1986) performed a meta-analysis of studies focusing on families and found several factors, such as a lack of parental supervision, inconsistent discipline, and low parent–child attachment, to be predictive of delinquency (Ayers et al., 1999; Dishion, Capaldi, & Yoerger, 1999). From a protective perspective, Henry, Tolan, and Gorman-Smith (2001) found that youth from families marked by closeness and strong parenting generally had lower levels of delinquency than those from families lacking cohesion and strong parenting. Peers become important as youths progress through adolescence and naturally spend more time outside the home building new relationships (Giordano, 2003; Smetana, Campione-Barr, & Metzger, 2006) and, although there is some question about the precise nature of the relationship (Haynie & Osgood, 2005), studies generally suggest that peers play a central role in antisocial behavior (Albert & Steinberg, 2011; Warr, 2002).

Moving into the broader social context, recent decades have seen increased attention to understanding the impact of communities on youth development (Leventhal, Dupéré, & Brooks-Gunn, 2009; Sampson, 2006; Wikström & Sampson, 2003). Among other problems, children growing up in disadvantaged, socially ineffectual neighborhoods may be at elevated risk of substance use and delinquency (Hawkins, Catalano, & Miller, 1992; Loeber & Wikström, 1993). Historically, at the community level, adolescent antisocial behavior has been studied from social organization, structural, and cultural perspectives (see review in Bursik & Grasmick, 1993), but, more recently, researchers have expanded those views to consider the active role that neighborhood collective efficacy may play in youth development (Sampson, 2006). There have, however, been mixed findings with respect to whether community-level factors influence trends in antisocial behavior over time (see, e.g., Loeber, Pardini, Stouthamer-Loeber, & Raine, 2007).

Current Study

The review points to the developmental interdependence among youth, their immediate social environment, and communities in the onset and continuance of antisocial behavior. Despite the value in using an inclusive framework, to this point studies have not fully considered how the qualities of the individual, immediate social environment, and community may affect antisocial behavior over time (Wikström & Sampson, 2003). So, while the need for holistic, multisystem intervention is acknowledged and promoted (Henggeler, 1997; Loeber & Farrington, 2000), the multifaceted process by which antisocial behavior emerges and continues requires further elaboration (Rutter, 1994). Farrington (2005), for example, notes that little is known about early risk factors in terms of the processes by which they give rise to long-term trends in antisocial behavior. The factors that foster risk or protection for antisocial behavior can be integrated with a “launch” framework to provide one means of looking at these processes.

This study examines influences on the initial observed level of antisocial behavior, reflecting an immediate effect, and the potential “launch” of a behavioral trajectory, which captures an enduring effect. Specifically, the conditional latent growth curve modeling approach used here provides estimates of the presence and degree of behavioral change/stability and whether this is affected by initial between-youth differences. The analysis first considers how to best characterize trajectories of substance use and delinquency across adolescence. This entails an assessment of initial levels (Intercept) and trends (Slope) and their variance estimates. Second, the study examines the extent to which key influences impact the initial level of substance use/delinquency. Third, the question of the extent to which key influences impact the progression (slope) of substance use/delinquency is examined. Fourth, the analysis considers whether youth trajectories of substance use and delinquency vary across neighborhoods. This consists of an (a) assessment of neighborhood cluster-level variance components for the Intercept and Slope and (b) exploration of neighborhood-level influences on the aggregated growth factors. Together, these different components of the analysis allow for consideration of what the typical growth pattern looks like in this sample and then follows with an elaborated model comprising several domains that have been found to be relevant in prior literature on adolescent antisocial behavior.

Data and Method

Data

Data from the Project on Human Development in Chicago Neighborhoods (PHDCN) were analyzed using latent growth curve models to study substance use and delinquency across three waves for three age cohorts (n = 752 [Cohort 9]; 752 [Cohort 12]; 626 [Cohort 15]). Each of the cohorts was interviewed approximately 2 years apart starting at the age used to label that group (e.g., Cohort 9 youths were observed at ages 9, 11, and 13). The approximate time of observation for the sample as a whole covered ages 9 through 19. Although these analyses could also be conducted using age to structure the analyses, recent research has found time and age approaches to reveal similar conclusions in relating covariates to longitudinal growth in antisocial behavior (Piquero, Monahan, Glasheen, Schubert, & Mulvey, 2012). Covariates were measured at Wave 1 for each cohort.

Units of observation were selected based on a multistage design where a random sample of 343 neighborhood clusters was initially chosen. Eighty of these clusters were then selected based on a stratified sampling strategy that focused on socioeconomic and racial composition. The selection of participants for the longitudinal cohort study followed from that process. Specifically, households and individual youth were randomly chosen from that set of neighborhoods and in-home, face-to-face standardized interviews were conducted with the youth and a primary caregiver (Earls & Visher, 1997). Seventy-five percent of those in Cohort 9, 74% in Cohort 12, and 72% in Cohort 15 who were invited to participate in the longitudinal cohort study actually did so (Molnar, Cerda, Roberts, & Buka, 2008). Retention levels for the cohorts included here ranged from 83 to 86% at Wave 2 and 71 to 78% at Wave 3. The analysis was expanded to the neighborhood level (N = 78) through use of data from a community survey. Approximately 8,700 adult residents (25–50 per neighborhood) were surveyed separately from the longitudinal cohort study regarding their perception of their neighborhood; the response rate was 78% within the neighborhood clusters (Earls & Visher, 1997; Liberman, 2007).

Study Measures

Outcomes

Self-reported Substance Use and Delinquency were measured at three waves for each cohort. Both sets of items refer to the frequency of behavior in the last year. A shared set of substance use and delinquency items was utilized to ensure similar coverage across waves and cohorts. The substance use measure comprises the number of self-reports of use for the previous year for alcohol, cocaine, marijuana, and other illicit drugs (e.g., heroin, amphetamines; National Institute on Drug Abuse, 1991). This is a single summed measure for the four forms of use mentioned here. Due to the prevalence of values at the lower end of the distribution (i.e., “0”) and skewness, these scores were log-transformed and treated as continuous, but censored, in the analysis (see Osgood, Finken, & McMorris, 2002). Self-reported delinquency was measured across three dimensions: violent offenses (e.g., attack with weapon, used weapon to get money/things), property offenses (e.g., damage property, stole from store), and public order/status offenses (e.g., skipped school). Specific items were chosen to fit with previous work on self-reported offending that has used PHDCN data as well as studies that have utilized this scale in the past (Huizinga, Esbensen, & Weihar, 1991). As was the case with substance use, the delinquency measures were log-transformed and treated as censored in the latent growth curve analysis.

Covariates

Individual Self-Control

A low self-control measure was developed based on the Emotionality, Activity, Sociability, and Impulsivity (EASI) temperament instrument (Buss & Plomin, 1975). Sixteen items from subscales tapping impulsivity (e.g., “trouble controlling impulses”), inhibitory control (e.g., “says first thing that comes into head”), sensation seeking (e.g., “seeks new, exciting experiences”), and persistence (e.g., “likes to see things through”) were used to create an additive score. The score was then standardized as in some prior PHDCN studies (see, e.g., Gibson, Sullivan, Jones, & Piquero, 2010; Cronbach’s α = .66); higher values reflect lower levels of self-control.

Family/Parental Influences

A series of measures that tap into the youth’s family environment and parental management processes are included in the analysis. The items were drawn from the Home Observation for Measurement of the Environment (HOME) Inventory (Caldwell & Bradley, 1984; Leventhal et al., 2004). The instrument is designed to assess the family environment as it contributes to youth development. Parental warmth is measured using 9 items derived from researcher observations of parents during the in-home interviews of primary caregivers and children (e.g., “parent encourages child to contribute”). Parental lack of hostility consists of items like “parent does not shout at child during visit” and “parent does not express annoyance with child.” All items were coded “0” (no) or “1” (yes) and individual responses were summed to create overall scores for both the parental warmth and the parental lack of hostility measures. Higher scores reflect more parental warmth and less hostility. Each of the scales exhibits acceptable reliability (α = .75 to .89), first factor explained variance (34 to 75%), and individual factor loadings (.39 to .87). Parental monitoring and supervision is a 13-item scale where primary caregivers self-reported on how they directly and indirectly monitor their children. This included items like “youth has a set time (curfew) to be home on weekend nights” and “establishes rules for behavior with peers and asks questions to determine whether they are being followed.” Higher scores on this scale indicate more supervision. Variance explained for the first extracted component (∼13 to 17%), factor loading values (.08 to .72), and reliability statistics (α = .43 to .57) are low for this measure.

Social Support

The Provision of Social Relationships (PSR) instrument asks questions about the degree to which the youth feels respected and has people (family and friends) whom they can count on if necessary (Turner, Frankel, & Levin, 1983). The three subscales of social support from family members, peers, and others (teachers and coaches) load highly on a single factor that explains 52% of the shared variance among the items. The factor loadings ranged from .44 to .69 and Cronbach’s α was .51. Items relevant to family members included whether the youth feels that his or her family “has confidence in them” or “helps them find solutions to problems.” Youth were also asked about their peer relationships in this regard. The “other social support” questions asked whether the youth had nonrelated adults that they could go to for help if needed (e.g., coaches, teachers).

Antisocial Peers

The peer influence measure comes from a set of 15 items that ask about the degree to which a youth’s “friends or people [they] spend time with” engage in delinquent activities and substance use (e.g., “number who purposely damaged property,” “number who attacked someone with weapon;” Huizinga, Esbensen, & Weihar, 1991). All items use the previous year as a reference period. The counts were summed and then averaged for each individual. Factor analysis indicated that a single factor explained about 35% of the variance in the underlying item covariance with individual loadings ranging from .42 to .69 (α = .86).

Sociodemographics

In addition to conditioning on age cohort (9, 12, and 15), three key control variables are utilized in the multivariate models. Gender was coded “0” for male and “1” for female and Race was coded “1” for White and “0” for non-White. The expanded racial distribution of the sample was as follows: 14.2% White, 35.9% Black, 45.9% Hispanic, and 3.9% other race. Socioeconomic status was constructed by PHDCN investigators and consisted of the principal component of parental caregiver’s (a) maximum education level (less than high school to bachelor’s degree or more), (b) salary (<$5,000 to >$50,000), and the maximum Socioeconomic Index occupational prestige of the parental caregiver and partner’s jobs. Forty-five percent of parental caregivers had less than a high school degree and only 9% had a bachelor’s degree or more. Approximately 21% of the sample had an annual primary earner income of less than $10,000 and 84% earned $50,000 or less.

Neighborhood Influences

Consistent with previous analysis of PHDCN data, the main measure used at the community level was Collective efficacy (see Sampson, Raudenbush, & Earls, 1997). Specifically, five social cohesion (e.g., “people around here are willing to help their neighbors”) and five neighborhood social control items (e.g., “neighbors would break up a fight in front of your house”) were combined to create the collective efficacy measure (α = .72). Additionally, two other neighborhood-level composite measures were included in neighborhood-level analyses: Social Capital (e.g., “there are adults in this neighborhood kids can look up to”), which consisted of 6 items (α = .88), and Social Disorder (e.g., “how much of a problem is groups of teenagers hanging out and causing trouble”), which consisted of five items (α = .78).

Analytic Plan

The analysis proceeds in a few steps. All models were estimated in MPlus 6.1 (Muthén & Muthén, 1998–2010) with full information maximum likelihood (Schafer & Graham, 2002). The unconditional models provide a general description of trajectories of adolescent antisocial behavior in this sample. Their key estimates comprise the initial level of the behavior in question (Intercept) as well as its rate and direction of change over time (Slope). Five key parameters are estimated: Means for the Intercept and Slope; their respective Variances; and their Covariance (Lawrence & Hancock, 1998; Singer & Willett, 2003). The intercept–slope Covariance estimate reflects the relationship between the initial level of antisocial behavior and its trend over time.

In addition to establishing a baseline of the sample-average longitudinal patterns and associated variation, the unconditional models inform a set of analyses incorporating covariates. Covariates capturing between-individual differences in the youth’s disposition, family socialization and monitoring practices, and peer delinquency at the initial wave of data collection were incorporated into a latent variable regression model to determine whether they had an effect on the growth factors. Specification of this model captures the effects of individual propensity and social influences on the trajectories. Both were expected to explain some variation in the estimated growth curves. In the final phase of the analysis, the models were specified in a multilevel framework to (a) assess the degree to which latent slopes and intercepts (i.e., growth parameters) vary across neighborhood clusters and (b) consider potential sources of that variation should it arise (Raudenbush & Bryk, 2002). Specifically, the latent growth factors were free to vary across neighborhood cluster, which captures the possibility that, collectively, youths from a given neighborhood may have significantly different trajectories of antisocial behavior than their peers from other areas of the city. This part of the analysis also considers whether aggregate community-level trajectories may be influenced by covariates at that level.

Results

Descriptives

Table 1 shows the descriptive statistics for the outcome measures and covariates. There are some cohort differences in the level of antisocial behavior at each particular wave, which is not surprising given the age trends that are typically observed in these behaviors (see Farrington, 1986). All differences were statistically significant with Kruskal–Wallis tests.

Table 1.

Descriptive Statistics for Key Covariates and Outcome Measures by Cohort Group.

	Full Sample	Cohort 9	Cohort 12	Cohort 15
	M (SD)	M (SD)	M (SD)	M (SD)
	Range	n = 752	n = 752	n = 626
Outcome measures
Substance use W1 (log)	0.20 (0.52)	0.002 (0.03)	0.06 (0.26)	0.60 (0.77)
Substance use W1 (log)	0–3.18	0.002 (0.03)	0.06 (0.26)	0.60 (0.77)
Substance use W2 (log)	0.48 (0.80)	0.03 (0.18)	0.35 (0.63)	1.20 (0.94)
Substance use W2 (log)	0–3.37	0.03 (0.18)	0.35 (0.63)	1.20 (0.94)
Substance use W3 (log)	0.86 (0.94)	0.24 (0.51)	0.93 (0.90)	1.55 (0.89)
Substance use W3 (log)	0–3.43	0.24 (0.51)	0.93 (0.90)	1.55 (0.89)
Delinquency W1 (log)	0.85 (1.29)	0.41 (0.96)	0.77 (1.16)	1.48 (1.53)
Delinquency W1 (log)	0–7.15	0.41 (0.96)	0.77 (1.16)	1.48 (1.53)
Delinquency W2 (log)	0.73 (1.22)	0.39 (0.87)	0.76 (1.09)	1.10 (1.56)
Delinquency W2 (log)	0–7.25	0.39 (0.87)	0.76 (1.09)	1.10 (1.56)
Delinquency W3 (log)	0.66 (1.26)	0.45 (0.92)	0.67 (1.26)	0.90 (1.55)
Delinquency W3 (log)	0–6.77	0.45 (0.92)	0.67 (1.26)	0.90 (1.55)
Covariates
Race (White = 1)	0.14	0.14	0.14	0.15
Gender (Male = 1)	0.50	0.47	0.51	0.51
SES (Factor score)	−0.12 (1.40)	−0.09 (1.41)	−0.17 (1.38)	−0.09 (1.42)
SES (Factor score)	−3.16–3.52	−0.09 (1.41)	−0.17 (1.38)	−0.09 (1.42)
Self-Control scale (Raw)	42.93 (10.91)	43.7 (10.3)	42.2 (10.6)	42.9 (11.8)
Self-Control scale (Raw)	0–80.0	43.7 (10.3)	42.2 (10.6)	42.9 (11.8)
Parental lack hostility	3.71 (.88)	3.66 (.96)	3.77 (.76)	3.70 (.89)
Parental lack hostility	0–4.00	3.66 (.96)	3.77 (.76)	3.70 (.89)
Parental warmth	6.36 (2.15)	6.49 (2.15)	6.51 (2.06)	6.01 (2.23)
Parental warmth	0–9.00	6.49 (2.15)	6.51 (2.06)	6.01 (2.23)
Parental monitoring	11.63 (1.51)	11.87 (1.28)	11.74 (1.52)	11.20 (1.67)
Parental monitoring	2.00–13.00	11.87 (1.28)	11.74 (1.52)	11.20 (1.67)
Total social support	48.05 (5.73)	47.80 (6.21)	48.71 (5.21)	47.55 (5.66)
Total social support	6.00–57.00	47.80 (6.21)	48.71 (5.21)	47.55 (5.66)
Antisocial peers	1.32 (0.29)	1.23 (0.24)	1.27 (.25)	1.48 (.33)
Antisocial peers	0.13–2.60	1.23 (0.24)	1.27 (.25)	1.48 (.33)

Latent Growth Curve Models

Estimation of the unconditional latent growth curve models was the initial step in the main analysis. The possibility of pooling the three cohorts in the latent growth curve models to capture lengthier behavioral trends was investigated before moving forward with that analysis (see Duncan, Duncan, & Strycker, 2006; Raudenbush & Chan, 1992). The process entails specifying a model that allows growth parameters to be freely estimated across cohorts and a second that constrains them to be equal. The test, which follows a Chi-Square distribution, compares the log-likelihood values for the free and constrained models. Differences in the log-likelihood values between the free and constrained models were calculated and then adjusted by a correction factor associated with the estimator used here (Asparouhov & Muthén, 2010). The degrees of freedom for the test is the difference in the number of free parameters for the models (10), leading to a critical value of 25.2 for rejection of the null hypothesis favoring the constrained model (p < .05). With model difference test values of 461.1 (delinquency) and 1,155.7 (substance use), the null hypothesis was easily rejected in both comparisons. This indicates that the additional free parameters in the full model are useful in fitting the data, and, consequently, the cohorts cannot be pooled to follow a single trajectory.

The unconditional latent growth curve results are presented in the top panel of Tables 2 and 3. On average, the estimated initial substance use score gets progressively higher from Cohort 9 (−7.81) to Cohort 15 (0.03). The slope is positive for all three cohorts, but the rate of linear growth is lower in Cohort 15 (0.37) than in the others (1.56, 0.79 for Cohorts 9 and 12, respectively). There is significant variation in both the initial level of substance use and its slope for Cohorts 12 and 15. The covariance between the intercept and slope is statistically significant and negative for those cohorts as well. This suggests that youth who start out with higher levels of substance use at Wave 1 tend to increase more slowly in subsequent years. As shown in Figure 1, substance use is low and fairly stable for the first two waves of Cohort 9; there is a subsequent increase from ages 11 to 13. This upward trend is picked up in the latter two waves of Cohort 12 and continues through the balance of the observation period for Cohort 15.

Figure 1.

Unconditional growth curves trends by cohort across age 9 to 19.

Table 2.

Latent Growth Curve Models With Covariates for Substance Use (Censored).

	Cohort 9	Cohort 12	Cohort 15
	Est (SE)	Est (SE)	Est (SE)
	n = 752	n = 752	n = 626
Estimates—unconditional model
Intercept	−7.81 (2.25)*	−2.64 (0.30)*	0.03 (0.07)
Slope	1.56 (0.56)*	0.79 (0.08)*	0.37 (0.02)*
Intercept variance	0.99 (5.90)	2.49 (0.51)*	1.69 (0.20)*
Slope variance	0.12 (0.18)	0.19 (0.03)*	0.09 (0.02)*
Intercept−Slope Covariance	0.11 (0.34)	−0.47 (0.10)*	−0.27 (0.05)*
Estimates—conditional model
Covariates Effects-Intercept
Race (White = 1)	0.99 (1.19)	0.40 (0.28)	0.55 (0.14)*
Gender (Female = 1)	0.73 (0.77)	−0.17 (0.21)	−0.10 (0.11)
SES (Factor score)	0.26 (0.36)	0.05 (0.08)	−0.02 (0.04)
Self-Control scale	0.25 (0.45)	−0.03 (0.10)	0.10 (0.05)⁁
Parental lack hostility	−0.05 (0.34)	−0.25 (0.14)⁁	0.06 (0.01)
Parental warmth	−0.12 (0.15)	0.05 (0.05)	−0.03 (0.02)
Parental monitoring	0.40 (0.37)	−0.04 (0.06)	0.003 (0.03)
Total social support	−0.03 (0.06)	0.00 (0.02)	0.001 (0.01)
Antisocial peers	1.33 (1.66)	2.34 (0.40)*	1.96 (0.24)*
Covariates effects-slope
Race (White = 1)	−0.17 (0.30)	−0.09 (0.07)	−0.03 (0.04)
Gender (Female = 1)	−0.18 (0.20)	0.00 (0.05)	−0.07 (0.03)*
SES (Factor Score)	−0.06 (0.09)	0.002 (0.02)	0.02 (0.01)
Self-Control scale	−0.07 (0.12)	0.03 (0.03)	−0.001 (0.02)
Parental lack hostility	0.03 (0.04)	0.05 (0.03)	−0.02 (0.02)
Parental warmth	0.03 (0.04)	−0.01 (0.01)	0.02 (0.01)*
Parental monitoring	−0.12 (0.09)	−0.01 (0.01)	−0.02 (0.01)⁁
Total social support	0.01 (0.02)	0.003 (0.01)	0.002 (0.003)
Antisocial peers	−0.13 (0.43)	−0.54 (0.09)*	−0.34 (0.05)*
Key estimates—multilevel model
Unconditional model
Within NH Variance-Intercept	0.63 (2.25)	0.43 (0.20)*	1.44 (0.09)*
Within NH Variance-Slope	0.14 (0.13)	0.17 (0.02)*	0.07 (0.01)*
Between NH Variance-Intercept	0.04 (0.20)	0.29 (0.14)*	0.004 (0.01)
Between NH Variance-Slope	0.004 (0.02)	0.01 (0.01)⁁	0.001 (0.01)

*p < .05. ⁁p < .10.

Table 3.

Latent Growth Curve Models With Covariates for Logged Delinquency (Censored).

	Cohort 9	Cohort 12	Cohort 15
	Est (SE)	Est (SE)	Est (SE)
	n = 752	n = 752	n = 626
Estimates—unconditional model
Intercept	−2.32 (0.22)*	−0.23 (0.10)*	1.06 (0.09)*
Slope	0.21 (0.07)*	−0.13 (0.04)*	−0.51 (0.05)*
Intercept variance	4.57 (1.01)*	1.88 (0.49)*	2.85 (0.58)*
Slope variance	0.09 (0.11)	0.07 (0.07)	0.26 (0.10)*
Intercept–slope covariance	−0.32 (0.28)	0.17 (0.15)	0.14 (0.15)
Estimates—Conditional model
Covariates effects-Intercept
Race (White = 1)	0.60 (0.45)	−0.24 (0.26)	−0.08 (0.24)
Gender (Female = 1)	−0.74 (0.26)*	0.06 (0.16)	−0.07 (0.15)
SES (Factor score)	0.11 (0.10)	0.08 (0.06)	0.11 (0.06)⁁
Self-Control scale	0.40 (0.15)*	0.23 (0.08)*	0.21 (0.07)*
Parental lack hostility	0.13 (0.16)	0.02 (0.11)	−0.06 (0.09)
Parental warmth	0.06 (0.07)	−0.08 (0.04)*	−0.02 (0.03)
Parental monitoring	−0.05 (0.09)	−0.04 (0.03)	0.03 (0.05)
Total social support	−0.02 (0.02)	−0.03 (0.02)*	−0.02 (0.02)
Antisocial peers	3.30 (0.69)*	3.34 (0.37)*	3.18 (0.26)*
Covariates effects-Slope
Race (White = 1)	−0.18 (0.13)	−0.07 (0.12)	−0.05 (0.13)
Gender (Female = 1)	0.08 (0.09)	−0.24 (0.07)*	−0.25 (0.08)*
SES (Factor score)	−0.03 (0.03)	−0.04 (0.03)	0.01 (0.03)
Self-Control scale	0.02 (0.05)	0.06 (0.04)⁁	0.01 (0.04)
Parental lack hostility	−0.02 (0.04)	−0.08 (0.05)⁁	0.01 (0.05)
Parental warmth	−0.01 (0.02)	0.03 (0.02)*	0.01 (0.02)
Parental monitoring	−0.01 (0.04)	−0.01 (0.02)	−0.05 (0.03)⁁
Total social support	0.004 (0.01)	0.02 (0.01)*	0.02 (0.01)*
Antisocial peers	−0.39 (0.20)*	−0.21 (0.14)	0.07 (0.14)
Key estimates—multilevel model
Unconditional model
Within NH Variance-Intercept	70.2 (9.2)*	1.46 (0.20)*	1.43 (0.26)*
Within NH Variance-Slope	3.28 (0.65)*	0.09 (0.03)*	0.18 (0.06)*
Between NH Variance-Intercept	0.31 (0.31)	0.29 (0.12)*	0.18 (0.13)
Between NH Variance-Slope	0.00 (0.001)	0.01 (0.01)	0.02 (0.06)

*p < .05. ⁁p < .10.

The conditional substance use estimates, which reflect expected changes in the initial level of substance use or its slope over time, differ somewhat across the three cohorts in terms of the consistency of effects. There were no significant effects for the Cohort 9 substance use growth curve. Parental lack of hostility has a marginally significant, protective effect on the intercept for Cohort 12 (−0.25). Exposure to delinquent peers has a significant effect in Cohorts 12 and 15. Specifically, a one-unit increase in the level of exposure to antisocial peers predicts higher values for substance use at the initial observation point for Cohorts 12 (b = 2.34) and 15 (b = 1.96). The effects on their respective slopes indicate that increased antisocial peer relationships at Wave 1 are associated with greater stability in the trend over time (b = −0.54, −0.34). The race measure has a significant effect on the substance use intercept for Cohort 15. White youth are expected to have higher levels of substance use at Wave 1 compared to youth from other races (b = 0.55). Only one self-control estimate reached marginal significance in this set of models (Cohort 15 Intercept).

The results for the unconditional latent growth curve models for delinquency are somewhat different than those for substance use. For all cohorts, each of the mean estimates for the intercept and slope is statistically significant. The initial estimated values get progressively higher in each of the three cohorts (Cohort 9 = −2.32, Cohort 12 = −0.23, Cohort 15 = 1.06). The estimated slope value is positive for Cohort 9 (0.21), but negative for Cohort 12 (−0.13) and Cohort 15 (−0.51). This is the key difference in the substance use and delinquency trends. Across the three waves, the trends for the delinquency scores indicate a slight increase for Cohort 9 and a slight decline for Cohort 12 (see Figure 1), reflecting the downward trend in antisocial behavior apparent in the later stages of adolescence. The mean scores at Wave 1 for Cohort 15 are much higher than those observed for any other cohort or measurement wave. The fact that the cohorts do not seamlessly overlap during common time points illustrates between-cohort differences in delinquency curves. There is significant variation in the intercepts for each of the three cohorts but, except for the Cohort 15 slope, the slope variance and intercept–slope covariance values are not statistically significant in this model.

Only self-control and exposure to delinquent peers were uniformly significant in their effects on the initial level of delinquency across cohorts. The results suggest a fairly sizable effect across cohorts where a unit increase in the delinquent peer measure was associated with a roughly 3-point increase on the latent intercept (b = 3.18–3.34). The self-control measure also has a significant effect for Cohort 9 (b = 0.40), Cohort 12 (b = 0.23), and Cohort 15 (b = 0.21). Specifically, youth with higher values (reflecting a lack of self-control) tend to also have higher initial levels of delinquency. Both parental warmth (b = −0.08) and social support (b = −0.03) had significant protective effects on the initial level of delinquency in the Cohort 12 analysis. The gender estimate for Cohort 9 (b = −0.74) indicates that females tend to start off with lower delinquency scores than males. Some of the same measures emerge as statistically significant in examining the estimates from the regression of the latent slope (longitudinal trend) on the covariates. Antisocial peers, parental warmth, and social support show effects that trend in the opposite direction from their relationships with the intercepts. The Gender estimates are also significant, suggesting that, for Cohorts 12 and 15, females have steeper downward slopes than males. For antisocial peers, the effect on the slope for Cohort 9 is negative (b = −0.39), which reflects greater stability over time for those individuals that have a greater volume of delinquent peers initially. Positive effects for the parental warmth and social support variables indicate that youth with higher scores tend to experience less steep declines in delinquency over time.

Multilevel Models

Multilevel growth curve models were fit to consider the possible role of the broader neighborhood context in understanding variation in longitudinal patterns of antisocial behavior (see bottom panels of Tables 2 and 3). Given the limited number of youth in some neighborhoods and the limited variation in several growth factors, this analysis is somewhat exploratory. Nevertheless, it reveals limited neighborhood-level differences in the aggregate growth trends. First, for Cohort 12, it appears that there is significant neighborhood-level variation in the initial level of substance use (0.29). One of the six growth estimates for delinquency offenses varies significantly at the neighborhood level as well: Cohort 12 intercept (0.29). The next stage of the analysis considered whether any of the three aspects of social structure or process could help explain between-person variance in the growth factors, controlling for the measures discussed above (results not shown but available upon request). The most relevant factor from the standpoint of previous research and theory was “collective efficacy.” Given the high correlations of collective efficacy and some of the other measures (e.g., Social capital), both single predictor and multivariate neighborhood-level models were considered. The collective efficacy measure was only significant in relation to the Cohort 12 delinquency intercept (−0.63). Other PHDCN neighborhood-level covariates such as social disorder and social capital failed to show expected effects on the neighborhood-level growth factors.

Discussion

Summary of Key Findings

This analysis of the PHDCN data, which are comprised of youth who are predominantly minority and relatively disadvantaged socioeconomically, yielded several important findings related to the conceptual framework described earlier. First, the patterns of behavioral trajectories in the unconditional models both fit with expectations based on previous research and offer some departures. Second, cohort differences in initial levels and longitudinal trends in antisocial behavior as well as relationships with covariates were identified. Third, exposure to delinquent peers affects the initial level of antisocial behavior in most cases and also has some effects on developmental trends later in adolescence. Fourth, individual self-control had a significant effect on the initial level of delinquency, but limited effects on its later trend or on substance use. Fifth, although the analyses identify some family influence effects, these are inconsistent across cohorts and outcome measures and appear to be somewhat limited in impact. Sixth, there are relatively few significant predictors in the conditional latent growth models. This is likely attributable in part to the limited variation in some of the latent growth factors. Seventh, the results from the unconditional multilevel growth curve models suggest that, in some tests, variation in the estimated growth factors was statistically significant across neighborhoods. Finally, in several cases, individual covariates had directional differences in their effects on initial levels and trends.

Implications for Developmental Understanding of Antisocial Behavior and Launch Model

The observed patterns in the delinquency measures mostly fit with expectations based on the age-crime curve in terms of a peak in mid-adolescence followed by a pronounced drop as youths move toward early adulthood. The longitudinal trend in substance use, which increases across the observation period, is clearly distinct from the trend in delinquency. The fact that alcohol was included in the substance use measure likely had some impact on that pattern, but it should nevertheless be kept in mind in considering the association (or possible lack thereof) between the two and what it means to desist/persist in delinquency versus substance use.

Individual self-control had a significant influence on the initial level of delinquency, but did not have an effect on its longer term trend. It also did not show any significant relationship with substance use. This suggests necessary caution in attempts to explain long-term trends using only between-individual differences in youth makeup. Peers were a robust influence on both the initial level and the longer term trends in these behaviors, but, in looking at the findings for other forms of social influence, the analysis of covariates identified relatively few instances of family or social support influence on immediate levels or long-term trends. Although methodologically due to limited sample sizes and variation in neighborhood growth curves the observed variation in the trajectories of antisocial behavior across neighborhood is important because it suggests potential community effects on development (Sampson, 2006; Wikström & Sampson, 2003).

Although the directional differences in effects on the initial levels of these behaviors and their subsequent trends appear to be counterintuitive at first glance (e.g., delinquent peers), they have implications for the launch model described earlier and, in turn, for understanding developmental patterns in these behaviors more generally. This reflects what Hussong and colleagues (2008) found when looking at substance use and criminal offending in the transition to adulthood; it signifies that a covariate can impact where a youth starts in terms of their behavior, as well as stability or change over time. So, for example, positive initial effects for association with antisocial peers suggest that those youth who report a greater number of such peers engaged in delinquency/substance use tend to have higher initial levels of delinquency or substance use. The negative effect on the slope suggests that youth could also change more or less markedly, depending on the observed direction of the latent slope and regression estimate, over time as well. For example, when a youth starts high or low initially she or he may get “locked into” a stable path over that time period. Alternately, she or he might have a higher positive or negative slope indicating something of a regression toward the mean. They may, however, still end up at a higher level of antisocial behavior compared to others in the sample. Similar patterns were revealed for the self-control measure, suggesting that a youth’s ability to regulate their behavior can have both immediate and long-term implications that should be assessed in analysis of risk. At the same time, collectively, the formal tests of the launch hypothesis showed limited effects for some common risk factors (e.g., family influences, SES) and inconsistent findings for others (e.g., self-control). Collectively, these tests suggest that initial risk does not necessarily embed a youth firmly in a given behavioral pattern.

Implications for Intervention

Although limited in some ways, the assessment of developmental trends in antisocial behavior using the approach taken here does offer some practical insights. In particular, the investigation of a “launch” perspective on antisocial behavioral development provides a sense of possible leverage points in designing prevention strategies for youth who are likely to be serious and sustained offenders as well as those whose antisocial behavior may be fleeting. First, given some identified differences across cohorts, the potential for differential risk/protection relationships by age should be considered in programming. Second, although it was somewhat surprising that self-control did not have effects as might be expected based on the underlying theory and previous findings, the results suggest that, on balance, individual propensity should be considered in pursuing other intervention strategies and appropriate measures should be taken to develop and use programming that has demonstrated effectiveness in bolstering skills related to self-regulation (Piquero, Jennings, & Farrington, 2010). This may be even more relevant in delinquency prevention, where the salience of self-control appears greater, than in substance use where trends may be driven more prominently by other factors (e.g., peers).

A number of best practices for at-risk youth are built around family-based programming. These initiatives are essential, but, given the general importance of peers in adolescents’ lives (Giordano, 2003) and their relationship to antisocial behavior (Warr, 2002), situational/peer risk should be a target of intervention as well. The robustness of delinquent peer exposure as a significant influence—even in the cohort observed starting in late childhood—points to the need for interventions that can counteract this salient risk factor or operationalize it as a protective factor (Sullivan & Jolliffe, 2012). Also, although the analysis is fairly limited, (see e. g., multi-level variance estimates for cohort 9 delinquency), it is apparent that not all of the variation in developmental trends in adolescent antisocial behavior is explained by individual factors or proximal social influences. This suggests that interventions that do not consider the community as a contributor to developmental processes may fall short in redirecting at-risk youth toward prosocial pathways. In particular, it may be useful to take the approach of the Communities that Care program where residents are asked to report on specific risks and needs of youth in the area, which in turn drives selection of best practices for prevention and intervention (Hawkins et al., 2008).

Limitations and Future Research

While these data have a number of strengths for investigating the development of antisocial behavior, there are also limitations that place boundaries around the conclusions. In particular, the data on covariate influences were limited to the first wave of observation for each cohort. This precluded testing for alternative or complementary explanations of the developmental patterns observed in these data. Use of time-varying covariates would further illuminate the factors that may have contributed to observed stability or change in behavioral trajectories. Additionally, some of the social influence covariates were relatively weak in their reliability values (e.g., parental warmth, social support) and/or were limited in other ways (e.g., measurement of parental monitoring and supervision may have benefited from youth reports). The measures of warmth and hostility in parenting are based on interviewer observations of parent–child interactions that are of a fairly short duration. Clearly, there may have been some social desirability effects at work in how the parent interacted with their child during that time period and this short duration surely would not fully encapsulate their relationship. Still, some researchers have made arguments in favor of this type of assessment (e.g., Patterson, 1982) and these measures have been found to be valid and reliable other analyses of the PHDCN data (Leventhal et al., 2004). The measure of exposure to antisocial peers used here comprises an individual’s reporting of their friend’s activities, which is limited in that they come from the same source as the delinquency measure (Haynie & Osgood, 2005; Zimmerman & Messner, 2011) and do not offer insight into the specific process by which peers affect individual behavior. Future studies should consider this in more depth—particularly as pertains to the interdependence among social influences.

Although the inclusion of the neighborhood indicator in the analysis allowed for some useful insight into individual developmental trends, there were clear measurement and analytic limitations at that level. For example, neighborhood census measures were not available in the restricted PHDCN data set used here. This is a limitation in terms of fully specifying distal processes at the neighborhood level that may affect outcomes considered here. Still, previous research indicates that the collective efficacy measure mediates the relationship between “social composition” variables, like concentrated disadvantage, in analysis of outcomes similar to those considered here (see Sampson, Raudenbush, & Earls, 1997). Additionally, the measures included in the community survey reflect the perceptions of adults in those communities rather than the youth (or those like them) included in the individual level analyses. Finally, the estimation of key effects became somewhat unstable once latent growth trends, cohort effects, controls for individual and immediate social covariate effects, and neighborhood-level influences were modeled together. Future research might disaggregate some of these tests to look at important questions regarding the factors that give rise to the trajectories of antisocial behavior. Additionally, the models do make assumptions about the nature of the growth trajectories and their variance. Nagin’s (2005) work on group-based trajectory models offers an alternative means of looking at the variation in longitudinal trends within a latent growth analysis framework. Further analyses with these data could consider the degree to which latent classes characterized by qualitatively different growth trends emerge and investigate their relationship to key individual and social influences. For instance, it is possible that a “launch” process may be more or less salient among certain developmental subgroups.

Conclusion

The understanding of adolescent antisocial behavior has increased markedly in recent years, to the point that recent legal opinions, such as the 2005 U.S. Supreme Court decision ruling on capital punishment for crimes committed as a juvenile in Roper vs. Simmons, and policy discussions, such as the possibility of reestablishing a higher boundary age for criminal court jurisdiction (Henrichson & Levshin, 2011), draw in part on the work of developmental researchers. Nevertheless, more can be done to pinpoint the etiology of antisocial behavioral trajectories. This study utilized data from the PHDCN to investigate longitudinal trends in adolescent delinquency and substance use along with key individual, social, and community influences. Assessing a launch model of antisocial behavioral trajectories, the study found significant variation in 6-year developmental trajectories in antisocial behavior across individuals and, in two cases, neighborhoods. Furthermore, some individual and social influences captured at the initial stage of PHDCN measurement were helpful in explaining this variation. These findings offered useful insights for understanding the processes that may give rise to trends in antisocial behavior in adolescence and associated prevention strategy.

Footnotes

Acknowledgments

The data used for this study were made available by the Inter-University Consortium of Political and Social Research (ICPSR). The author would like to acknowledge NIJ for support for participation at an ICPSR workshop on use of the PHDCN data set.

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.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by funding from the National Institute of Justice (09-IJ-CX-0042). The conclusions reported here are those of the author and do not reflect the opinions of the agency.

References

Albert

Steinberg

(2011). Peer influences on adolescent risk behavior. In Bardo

Fishbein

Milich

(Eds.), Inhibitory control and drug abuse prevention: From research to translation (pp. 211–226). New York, NY: Springer.

Asparouhov

Muthén

(2010). Computing the strictly positive Satorra-Bentler Chi-square test in MPlus. Retrieved from www.statmodel.com/examples/webnote.shtml#web12

Ayers

C. D.

Williams

J. H.

Hawkins

J. D.

Peterson

P. L.

Catalano

R. F.

Abbott

R. D.

(1999). Assessing correlates of onset, escalation, de-escalation, and desistance of delinquent behaviors. Journal of Quantitative Criminology, 15, 277–306.

Bursik

R. J.

Grasmick

H. G.

(1993). Neighborhoods and crime: The dimensions of effective community control. Lanham, MD: Lexington Books.

Buss

A. H.

Plomin

(1975). A temperament theory of personality development. New York, NY: John Wiley.

Caldwell

Bradley

(1984). Home observation for measurement of the environment. Little Rock: University of Arkansas.

Catalano

R. F.

(2007). Prevention is a sound public and private investment. Criminology and Public Policy, 6, 377–398.

Cohen

M. A.

Piquero

A. R.

(2009). New evidence on the monetary value of saving a high risk youth. Journal of Quantitative Criminology, 25, 25–49.

Dishion

T. J.

Capaldi

D. M.

Yoerger

(1999). Late childhood antecedents to progressions in male adolescent substance use: An ecological analysis of risk and protection. Journal of Adolescent Research, 14, 175–205.

10.

Duncan

S. C.

Duncan

T. E.

Strycker

L. A.

(2006). Alcohol use from ages 9-16: A cohort-sequential latent growth model. Drug and Alcohol Dependence, 81, 71–81.

11.

Earls

Visher

(1997). Project on human development in Chicago neighborhoods: a research update. Washington, DC: Office of Justice Programs, National Institute of Justice.

12.

Farrington

D. P.

(1986). Age and crime. In Tonry

(Ed.), Crime and justice (Vol. 7, pp. 189–250). Chicago, IL: University of Chicago Press.

13.

Farrington

D. P.

(2005). Introduction to integrated developmental and life-course theories of offending. In Farrington

D. P.

(Ed.), Advances in criminological theory: Integrated developmental and life-course theories of offending (Vol. 14, pp. 1–14). New Brunswick, NJ: Transaction.

14.

Farrington

D. P.

(2006). Building developmental and life-course theories of offending. In Cullen

F. T.

Wright

J. P.

Blevins (Eds.), Advances in criminological theory: Integrated developmental and life-course theories of offending (Vol. 15, pp. 335–364). New Brunswick, NJ: Transaction.

15.

Farrington

D. P.

Welsh

B. C.

(2007). Saving children from a life of crime: Early risk factors and effective interventions. New York, NY: Oxford University.

16.

Gibson

C. L.

Sullivan

C. J.

Jones

Piquero

(2010). “Does it take a village?” The role of neighborhood context in child and adolescent self control. Journal of Research in Crime and Delinquency, 47, 31–62.

17.

Giordano

P. C.

(2003). Relationships in adolescence. Annual Review of Sociology, 29, 257–281.

18.

Gottfredson

M. R.

Hirschi

(1990). A general theory of crime. Palo Alto, CA: Stanford University Press.

19.

Hawkins

J. D.

Catalano

R. F.

Arthur

M. W.

Egan

Brown

Abbott

R. D.

Murray

D. M.

(2008). Testing communities that care: The rationale, design and behavioral baseline equivalence of the community youth development study. Prevention Science, 9, 178–190.

20.

Hawkins

J. D.

Catalano

R. F.

Miller

J. Y.

(1992). Risk and protective factors for alcohol and other drug problems in adolescence and early adulthood: Implications for substance abuse prevention. Psychological Bulletin, 112, 64–105.

21.

Hawkins

J. D.

Herrenkohl

Farrington

D. P.

Brewer

Catalano

R. F.

Harachi

T. W.

(1998). A review of predictors of youth violence. In Loeber

Farrington

(Eds.), Serious & violent juvenile offenders: risk factors and successful interventions (pp. 106–146). Thousand Oaks, CA: Sage.

22.

Haynie

D. L.

Osgood

D. W.

(2005). Reconsidering peers and delinquency: How do peers matter? Social Forces, 84, 1109–1130.

23.

Henggeler

S. W.

(1997). Treating serious antisocial behavior in youth: The MST approach. Washington, DC: US Department of Justice, Office of Juvenile Justice and Delinquency Prevention.

24.

Henrichson

Levshin

(2011). Cost-benefit analysis of raising the age of juvenile jurisdiction in North Carolina. New York, NY: Vera Institute.

25.

Henry

D. B.

Tolan

P. H.

Gorman-Smith

(2001). Longitudinal family and peer group effects on violence and nonviolent delinquency. Journal of Clinical Child Psychology, 30, 172–186.

26.

Hser

Y. I.

Longshore

Anglin

M. D.

(2007). The life course perspective on drug use: A conceptual framework for understanding drug use trajectories. Evaluation Review, 31, 515–547.

27.

Huizinga

Esbensen

Weihar

A. W.

(1991). Are there multiple paths to delinquency? The Journal of Criminal Law and Criminology, 82, 83–118.

28.

Hussong

A. M.

Curran

P. J.

Moffitt

Caspi

(2008). Testing turning points using latent growth curve models: Competing models of substance abuse and desistance in young adulthood. In Cohen

(Ed.), Applied techniques for turning points research (pp. 81–104). New York, NY: Routledge.

29.

Jenson

J. M.

Fraser

M. W.

(2010). Social policy for children and families: A risk and resilience perspective (2nd ed.). Thousand Oaks, CA: Sage.

30.

Johnston

L. D.

O’Malley

P. M.

Bachman

J. G.

Schulenberg

J. E.

(2011). Monitoring the Future national results on adolescent drug use: Overview of key findings, 2010. Ann Arbor, MI: Institute for Social Research, The University of Michigan.

31.

Kinderman

T. A.

Skinner

E. A.

(1992). Modeling environmental development: Individual and contextual trajectories. In Basendorpf

Valsiner

(Eds.), Stability and change in development: A study of methodological reasoning (pp. 155–190). Newbury Park, CA: Sage.

32.

Knoll

Sickmund

(2010). Delinquency cases in juvenile courts, 2007. Washington, DC: U.S. Department of Justice, Office of Juvenile Justice and Delinquency Prevention.

33.

Lahey

Waldman

(2005). A developmental model of the propensity to offend during childhood and adolescence. In Farrington

D. P.

(Ed.), Advances in criminological theory: Integrated developmental & life-course theories of offending (Vol. 14, pp. 15–50). New Brunswick, NJ: Transaction.

34.

Lawrence

Hancock

(1998). Assessing change over time using latent growth modeling. Measurement and Evaluation in Counseling and Development, 30, 211–224.

35.

Leventhal

Dupéré

Brooks-Gunn

(2009). Neighborhood influences on adolescent development. In Steinberg

Lerner

(Eds.), Handbook of adolescent psychology (Vol. 2, pp. 411–443). New York, NY: John Wiley.

36.

Leventhal

Selmer-O’Hagan

M. B.

Brooks-Gunn

Bingenheimer

J. B.

Earls

F. J.

(2004). The homelife interview from the project on human development in Chicago neighborhoods: Assessment of parenting and home environment for 3- to 15-year-olds. Parenting, 4, 211–241.

37.

Liberman

(2007). Adolescents, neighborhoods, and violence: Recent findings from the Project on Human Development in Chicago Neighborhoods. Washington, DC: Office of Justice Programs, National Institute of Justice.

38.

Loeber

Farrington

D. P.

(2000). Young children who commit crime: Epidemiology, developmental origins, risk factors, early interventions, and policy implications. Development and Psychopathology, 12, 737–762.

39.

Loeber

Pardini

Stouthamer-Loeber

Raine

(2007). Do cognitive, physiological, and psychosocial risk and promotive factors predict desistance from delinquency in males? Development and Psychopathology, 19, 867–887.

40.

Loeber

Stouthamer-Loeber

(1986). Family factors as correlates and predictors of juvenile conduct problems and delinquency. In Tonry

Morris

(Eds.), Crime and Justice: An annual review of research (Vol. 7, pp. 29–149). Chicago, IL: University of Chicago Press.

41.

Loeber

Wikström

P. O. H.

(1993). Individual pathways to crime in different types of neighborhoods. In Farrington

D. P.

Sampson

R. J.

Wikström

P. O.

(Eds.), Integrating individual and ecological aspects of crime (pp. 169–204). Stockholm, Sweden: National Council for Crime Prevention.

42.

Molnar

B. E.

Cerda

Roberts

A. L.

Buka

S. L.

(2008). Effects of neighborhood resources on aggressive and delinquent behaviors among urban youths. American Journal of Public Health, 98, 1086–1093.

43.

Muthén

L. K.

Muthén

(1998-2010). Mplus User’s Guide (6th ed.). Los Angeles, CA: Author.

44.

Nagin

D. S.

(2005). Group-based modeling of development. Cambridge, MA: Harvard University.

45.

National Institute on Drug Abuse. (1991). National household survey on drug abuse: Population estimates,1991. Rockville, MD: Author.

46.

Osgood

D. W.

Finken

L. W.

McMorris

(2002b). Analyzing multiple-item measures of crime and deviance II: Tobit regression analysis of transformed scores. Journal of Quantitative Criminology, 18, 319–347.

47.

Patterson

G. R.

(1982). Coercive family process. Eugene, OR: Castalia.

48.

Piquero

A. R.

Jennings

Farrington

D. P.

(2010). On the malleability of self control: Theoretical and policy implications regarding a general theory of crime. Justice Quarterly, 27, 803–834.

49.

Piquero

A. R.

Monahan

Glasheen

Schubert, Mulvey

(2012). Does time matter? Comparing trajectory based concordance and covariate association using time-based and age-based assessments. Crime & Delinquency, 1-26. doi:10.1177/0011128712459491

50.

Raudenbush

S. W.

Bryk

A. S.

(2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Newbury Park, CA: Sage.

51.

Raudenbush

S. W.

Chan

W. S.

(1992). Growth curve analysis in accelerated longitudinal designs with application to the National Youth Survey. Journal of Research on Crime and Delinquency, 29, 387–411.

52.

Rutter

(1994). Beyond longitudinal data: Causes, consequences, changes, and continuity. Journal of Consulting and Clinical Psychology, 62, 928–940.

53.

Sampson

R. J.

(2006). How does community context matter? Social mechanisms and the explanation of crime. In Wikström

P. O.

Sampson

R. J.

(Eds.), The explanation of crime: Context, mechanisms, and development (pp. 31–60). New York, NY: Cambridge University Press.

54.

Sampson

R. J.

Laub

(1993). Crime in the making: Pathways and turning points through the life-course. Cambridge, MA: Harvard University Press.

55.

Sampson

R. J.

Raudenbush

S. W.

Earls

(1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277, 918–924.

56.

Schafer

J. L.

Graham

J. W.

(2002). Missing data: our view of the state of the art. Psychological Methods, 7, 147–177.

57.

Simons

R. L.

Simons

L. G.

Wallace

L. E.

(2004). Families, delinquency, and crime. New York, NY: Oxford University Press.

58.

Singer

J. D.

Willett

J. B.

(2003). Applied longitudinal data analysis: Modeling change and event occurrence. New York, NY: Oxford University.

59.

Smetana

J. G.

Campione-Barr

Metzger

(2006). Adolescent development in interpersonal and societal contexts. Annual Review of Psychology, 57, 255–284.

60.

Sullivan

C. J.

Jolliffe

(2012). Peer influence, mentoring and crime prevention. In Welsh

B. C.

Farrington

D. P.

(Eds.), Handbook of crime prevention. Oxford, UK: Oxford University Press.

61.

Turner

Frankel

Levin

(1983). Social support: Conceptualization, measurement, and implications for mental health. In Greeley

J. R.

(Ed.), Research in community and mental health (pp. 67–111). Greenwich, CT: JAI Press.

62.

Warr

(2002). Companions in crime: The social aspects of criminal conduct. Cambridge, England: Cambridge University.

63.

Welsh

B. C.

(2005). Public health and the prevention of juvenile criminal violence. Youth Violence and Juvenile Justice, 3, 23–40.

64.

Wikström

P. O. H.

Sampson

R. J.

(2003). Social mechanisms of community influences on crime and pathways in criminality. In Lahey

Moffitt

Caspi

(Eds.), The causes of conduct disorder and serious juvenile delinquency (pp.118–148). New York, NY: Guilford.

65.

Zimmerman

Messner

(2011). Neighborhood context and nonlinear peer effects on adolescent violent crime. Criminology, 49, 873–904.