Testing Moffit Drug Use Study

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

869190

research-article2019
CADXXX10.1177/0011128719869190Crime & DelinquencyWiddowson et al.

Article
Crime & Delinquency
2020, Vol. 66(3) 420­–445
Exposure to Persistently © The Author(s) 2019
Article reuse guidelines:
Delinquent Peers and sagepub.com/journals-permissions
DOI: 10.1177/0011128719869190
https://doi.org/10.1177/0011128719869190
Substance Use Onset: A journals.sagepub.com/home/cad

Test of Moffitt’s Social


Mimicry Hypothesis

Alex O. Widdowson1 , J. W. Andrew Ranson2,


Sonja E. Siennick3, Kelly L. Rulison4,
and D. Wayne Osgood5

Abstract
Moffitt’s social mimicry hypothesis states that the sudden rise in offending
during adolescence is partly the result of adolescence-limited delinquents
modeling the behavior of their life-course persistent peers. We test this
hypothesis using social network data from 7,742 adolescents followed from
6th to 12th grades to consider whether having a persistently delinquent
friend, especially one who used substances, predicted substance use
initiation. Results indicated that although having a persistently delinquent
friend was associated with an increased risk of general substance use
initiation, adolescents who had a persistently delinquent friend were just
as likely to initiate smoking, drinking, and marijuana use whether that friend
used that specific substance or not, which suggests that adolescents may not
mimic their friends’ use of specific substances.

1University of Louisville, KY, USA


2Marquette University, Milwaukee, WI, USA
3Florida State University, Tallahassee, FL, USA
4University of North Carolina at Greensboro, NC, USA
5Pennsylvania State University, University Park, PA, USA

Corresponding Author:
Alex O. Widdowson, Assistant Professor, Department of Criminal Justice, University of
Louisville, 2301 South Third St., Louisville, KY 40292, USA.
Email: alex.widdowson@louisville.edu
Widdowson et al. 421

Keywords
Moffitt, adolescent limited, peer influence, substance use

Over the past 25 years, numerous studies have tested different aspects of
Moffitt’s (1993) dual taxonomy theory, including the existence of different life-
course trajectories (e.g., Jennings & Reingle, 2012; Piquero, 2008) and the eti-
ology of the different groups (e.g., Barnes & Beaver, 2010; Piquero & Brezina,
2001; Raine et al., 2005; Tibbetts & Piquero, 1999). Despite this attention, no
study has directly tested one of the theory’s central hypotheses about the etiol-
ogy of adolescence-limited (AL) delinquency: the social mimicry hypothesis.
According to Moffitt (1993), the sudden rise in offending observed in adoles-
cence is the result of AL delinquents experiencing a “maturity gap” and model-
ing the behavior of their life-course persistent (LCP) peers to appear more
adult-like and independent. This modeling can occur in the context of adoles-
cents’ social networks. For example, Moffitt (1993) suggests that “[LCPs]
serve as core members of revolving networks, by virtue of being role models or
trainers of new recruits” (p. 688). In this study, we test this possibility by con-
ducting a social network analysis of the social mimicry hypothesis.
Although a few studies have examined the role that peers play in Moffitt’s
theory (e.g., Franken, Harakeh, Veenstra, Vollebergh, & Dijkstra, 2017;
Rulison, Kreager, & Osgood, 2014; Young, 2014), no study has examined the
extent to which having ties to LCP peers predicts the onset of delinquency
and substance use in adolescence. The lack of research testing this proposi-
tion is understandable given that few data sources include the design features
needed to directly examine this issue: (a) longitudinal social network data
during the period of adolescence when most youth initiate their criminal
careers and (b) a long enough period of observation to identify a group of
persistently delinquent youth that reasonably resembles LCP offending.
Our study addresses this void by testing Moffitt’s (1993) social mimicry
hypothesis by examining two research questions. First, does having a persis-
tently delinquent friend predict the onset of substance use? Second, does a
persistently delinquent friend’s own substance use predict adolescent sub-
stance use onset? Specifically, we assess whether the mimicry is specific to
the form of substance use (e.g., smoking by a persistently delinquent friend
predicts smoking by an AL offender) or whether mimicry is based on general
involvement in delinquency.

Background
Moffitt’s (1993) dual taxonomy is an important developmental theory in
criminology designed to explain the age-crime curve. According to Moffitt,
422 Crime & Delinquency 66(3)

the age-crime curve is not the same for all individuals; rather, it is an aggre-
gate curve concealing two distinct groups of offenders each with a unique
developmental course and etiology. The first group—LCP offenders—is
believed to make up 5% to 8% of the population, yet they are responsible for
the majority of criminal acts including the most serious and violent offenses.
As the name implies, LCP antisocial behavior spans the life course: It begins
in early childhood, perpetuates into adolescence, and persists well into adult-
hood. This group’s offending is said to be the result of neuropsychological
deficits and a disadvantaged rearing environment.
Numerous empirical studies have examined Moffitt’s claims about LCP
offenders (Moffitt, 2006). Results suggest that early neuropsychological defi-
cits are associated with greater levels of delinquency (Moffitt, Lynam, & Silva,
1994), that deficits and a disadvantaged environment interact to produce an
early onset of offending (Tibbetts & Piquero, 1999), and that neuropsychologi-
cal deficits predict whether an individual will become an LCP offender (Bellair,
McNulty, & Piquero, 2016). In addition, LCP offenders are also at risk for a
number of negative life outcomes in adulthood, including physical and mental
health problems, poor parenting practices, and employment difficulty (Moffitt,
Caspi, Harrington, & Milne, 2002; Odgers et al., 2008). Thus, there is good
support for Moffitt’s (1993) argument that there is a relatively small group of
offenders who initiate offending early in life and continue to offend well
beyond what one would expect from examining the age-crime curve.
The second group—called AL delinquents—is much larger, with the major-
ity of youth falling in this category. Moffitt argues that this group’s delin-
quency begins in adolescence and is a normal part of development. Under the
dual taxonomy, the majority of AL offenders will desist upon reaching adult-
hood so long as they do not become ensnared by the negative consequences of
antisocial behavior, such as a criminal record or dropping out of high school,
which limit or prevent successful transitions to adulthood.
Moffitt (1993) argues that AL delinquency is largely attributed to the
“maturity gap” and “social mimicry.” The maturity gap is the product of con-
temporary society’s restrictions on adolescents—that is, most adolescents are
biologically mature but are denied adult privileges (e.g., purchasing alcohol
and tobacco). This gap produces role conflict and strain among adolescents,
and they seek to reduce this dissonance by attempting to appear more adult-
like. Research generally supports the idea that a maturity gap leads to greater
levels of delinquency (Barnes & Beaver, 2010; Dijkstra et al., 2015; Piquero
& Brezina, 2001). In support of Moffitt’s (1993) argument that adolescent
delinquency is normative, research suggests that the maturity gap is associ-
ated with minor delinquency but unrelated to violence and aggression (Barnes
& Beaver, 2010; Piquero & Brezina, 2001). More recent research has also
investigated the pathways through which the maturity gap leads to
Widdowson et al. 423

delinquency (Dijkstra et al., 2015), showing that strain partially mediates the
relationship between the maturity gap and offending.
According to Moffitt’s (1993) theory, LCP offenders do not experience a
“maturity gap” and appear more mature (or adult-like) to AL offenders. For
example, LCP offenders shun adult norms and rules, they obtain possessions
through criminal behavior, they engage in substance use, and they are more
sexually experienced. Noticing that LCP offenders do not experience the
“maturity gap,” AL offenders begin to covet the behavior of LCP offenders,
leading to a surge in LCP popularity during adolescence among AL offenders
(i.e., the “role magnet” hypothesis). AL offenders then begin to model the
behavior of LCP offenders to feel and appear more adult-like and indepen-
dent (i.e., the “social mimicry” hypothesis). Thus, in addition to the more
frequently studied maturity gap, Moffitt (1993) also stipulates that “[AL]
delinquents begin delinquency” because of “social mimicry of the antisocial
lifestyle of [LCP] youths” (p. 686).
Moffitt suggests two mechanisms through which social mimicry can
occur. First, Moffitt (1993) suggests that modeling takes place in the context
of adolescents’ social networks where LCP offenders serve as “core mem-
bers” of a co-offending network—they are the recruiters, instigators, and
trainers—and AL offenders serve as “joiners” (p. 688). Under this concep-
tion, the friendship between LCP and AL offenders is one that is mutually
exploitative: LCP offenders exploit AL offenders as accomplices, lookouts,
fences, and drug clients, and, in turn, AL offenders learn the coveted delin-
quent lifestyle of LCP offenders. In support of this conception, Warr (1996)
found that the relationship between instigators and joiners in various crimes
was described as either best/very close friends or friends in the large majority
of cases (78%-98%). Second, Moffitt (1993) suggests that social mimicry can
also happen “from afar,” which would require no friendship ties (p. 688).
Moffitt notes that the same qualities that make LCP peers appear more adult-
like may also make them unsuitable as close friends (e.g., LCP peers lack
trust and loyalty). Under this conception, adolescents might observe an LCP
offender using substances at a party or after school despite not being friends,
or they might hear about such use through third-party communication. In this
study, we test the first mechanism by conducting a social network analysis of
the social mimicry hypothesis.
Although the social mimicry hypothesis is a key part of Moffitt’s theory, no
empirical work has directly tested this proposition (Moffitt, 2006; Piquero,
Diamond, Jennings, & Reingle, 2013). Despite the central role of LCP offend-
ers in AL offending, most work on AL offending focuses on the maturity gap
(e.g., Barnes & Beaver, 2010; Dijkstra et al., 2015; Piquero & Brezina, 2001).
Nevertheless, there is some indirect evidence bearing on the issue. Using data
424 Crime & Delinquency 66(3)

from the National Longitudinal Study of Adolescent Health (Add Health) and
Group-Based Trajectory Modeling (GBTM), Young (2014) identified three
groups of offenders: one that displayed chronic offending across adolescence
and into adulthood, one that displayed high but declining offending, and one
that displayed low levels of offending. Young then examined growth curves in
popularity. In line with Moffitt’s (1993) role magnet hypothesis, individuals in
the chronic offending group displayed overall lower popularity throughout the
study compared with other groups, but they experienced temporary within-
individual increases in popularity during adolescence with declines in early
adulthood. That is, LCP offenders’ popularity increased when their peers were
more likely to be experiencing the maturity gap but decreased as these peers
began to age out of delinquency and gain more autonomy. In another study,
Rulison and colleagues (2014) examined popularity trajectories among three
different groups of offenders using the PROmoting School–Community–
University Partnerships to Enhance Resilience (PROSPER) sample. Contrary
to Young’s (2014) study, the authors found that the popularity of persistent
delinquents did not increase from Grades 6 to 9. Importantly, this study focuses
on early and middle adolescence, which is the period that Moffitt argues should
be the apex of popularity for LCP offenders. Finally, using data from the Social
Network Analysis of Risk Behaviors study, Franken and colleagues (2016)
found that an early onset of externalizing behavior was associated with higher
perceived popularity among peers but lower likability, and the association
between early onset and number of friendships was not significant.
These studies test one aspect of Moffitt’s (1993) taxonomy—whether LCP
offenders become more popular during adolescence—but they do not directly
test the social mimicry hypothesis. That is, they do not test the impact of LCP
offenders on the behavior of AL offenders. The current body of literature
provides mixed indirect evidence in favor of the social mimicry hypothesis.
However, to our knowledge, this study is the first to examine the crux of the
social mimicry hypothesis—that is, the extent to which having ties to LCP
peers predicts the onset of deviant behavior in adolescence.

Current Study
The current study uses longitudinal survey and social network data from the
PROSPER study to test whether having a persistently delinquent friend is
associated with an increased risk of substance use onset from 6th to 12th
grades.

Research Question 1: Does having a persistently delinquent friend pre-


dict adolescent substance use initiation?
Widdowson et al. 425

If Moffitt’s theory is correct that exposure to an LCP friend is important for


the onset of deviant AL behavior, then having a persistently delinquent friend
should predict substance use onset.

Research Question 2: Does persistently delinquent friends’ own use pre-


dict adolescent substance use initiation?

If mimicry explains the onset of AL deviance, then AL youth should initiate


the specific behaviors that are displayed by their deviant friends. We examine
whether having a persistently delinquent friend who smokes, reports being
drunk, or uses marijuana predicts the onset of use of each corresponding sub-
stance. Although Moffitt did not explicitly require a substance-specific trans-
mission of social mimicry, modeling (or “mimicry”) of peers implies that
adolescents observe their friends’ behaviors (here, substance use) and initiate
those specific behaviors (here, the use of specific substances; Akers & Sellers,
2004, pp. 85-89). Indeed, the influence of peer substance use on adolescents’
use of the same substance has been documented in numerous studies, includ-
ing substance-specific effects of peer tobacco, alcohol, and marijuana use
(e.g., Bachman et al., 2002; Dinges & Oetting, 1993; Huba, Wingard, &
Bentler, 1979). This has been attributed to adolescent modeling or imitation
of peer behavior (Petraitis, Flay, & Miller, 1995).

Method
Data and Sample
We use data from the PROSPER study, a place-randomized substance abuse
prevention trial in 28 public school districts in rural Pennsylvania and Iowa
(Spoth, Greenberg, Bierman, & Redmond, 2004; Spoth et al., 2007).1 The
study included school districts from communities that had 45,000 residents or
less, were predominantly White (61%-96%), and had populations with at
least 15% of families eligible for free or reduced price school lunches.2
PROSPER’s sample consists of two successive cohorts of students who were
first interviewed in the fall of 6th grade with follow-ups conducted each
spring from 6th through 12th grade (a total of eight waves). An average of
10,000 students participated at each wave, with participation rates ranging
from 86% to 90% of students across waves. At each wave, students com-
pleted in-school surveys, which included items that assessed various aspects
of their development relevant to Moffitt’s theory, including their social lives
and involvement with substance use and delinquency, although the surveys
did not assess all aspects of Moffitt’s theory (e.g., the maturity gap).
426 Crime & Delinquency 66(3)

Like other social network studies of peer delinquency and substance use
(e.g., Add Health, Haynie, 2001; Netherlands Institute for the Study of Crime
and Law Enforcement [NSCR] School Project, Weerman & Smeenk, 2005),
the PROSPER study used respondents’ peer nominations to assess friendship
ties. At each wave, respondents nominated up to two best friends and five
close friends in their school and grade.3 The PROSPER staff was able to
match more than 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. As the entire
grade participated in the study, these data allowed us to construct complete
within-grade, school-based friendship networks. To create the peer measures
defined below, we linked respondents to their friends’ scores on relevant sur-
vey items, which allowed us to create “direct measures” of peer behavior that
avoid issues of perception and projection bias (Gottfredson & Hirschi, 1990;
Prinstein & Wang, 2005), although they do not allow us to determine whether
adolescents actually observed their friends’ behavior.
The analytical sample was restricted to respondents who had complete data
on delinquency in the fall of 6th grade (N = 10,789) and had complete data on
delinquency for at least four additional waves (N = 8,000). We made these
restrictions because we were interested in identifying a group of persistently
delinquent youth. By selecting respondents who were present in 6th grade (M
age = 11.8) and present for at least five total waves, we can identify respon-
dents who started delinquent behaviors at a relatively young age and who
display elevated and sustained delinquency across time. In addition, an analy-
sis of substance use initiation requires the repeated measurement of substance
use over time to accurately identify when respondents first initiated substance
use. We also excluded respondents who attended one Pennsylvania school dis-
trict that did not participate in the social network portion of the survey (N =
258). The final analytical sample consists of 7,742 respondents. To address
missing data on control variables, we used multiple imputation using chained
equations with the mim suite available in Stata. In doing so, we created 20
imputed data sets. Standard errors were calculated using Rubin’s (1987) rules,
which account for variance between and within the imputed data sets.
These restrictions allowed us to identify individuals who displayed ele-
vated delinquency over time, but the exclusion of some cases may introduce
bias into our analyses. We assessed this possibility by conducting a series of
t tests that compared our analytical sample with those lost due to attrition on
variables assessed at Wave 1. The excluded cases were more likely to be male
(t = 4.8), non-White (t = 10.1), from single-parent families (t = 14.4), and
free-lunch recipients (t = 21.5), and they had lower school grades (t = 21.4),
poorer family relations (t = 8.0), higher sensation seeking behavior (t = 7.7),
Widdowson et al. 427

and higher Wave 1 delinquency (t = 15.7). Given these differences, we con-


ducted sensitivity analyses that relaxed our exclusion criteria by using data
only from 6th to 9th grades.4 This strategy likely retained more LCP offend-
ers as those respondents would be more likely to miss later waves due to
dropping out of high school. This allowed us to retain 8,530 respondents. The
analyses based on the larger sample were consistent with the findings and
conclusions that we present below. Although our findings were robust to this
alternate selection criterion, attrition remains one limitation of our study.

Dependent Variables
Three outcome variables assessed respondents’ substance use initiation from
6th to 12th grades. At each wave, respondents reported whether they had ever
smoked a cigarette, been drunk from drinking alcohol, and smoked marijuana
or hashish. Using these items, we constructed wave-specific indicators of
whether respondents first reported using cigarettes (cigarette onset), being
drunk (drunkenness onset), or using marijuana (marijuana onset) at that
wave (0 = no, 1 = yes). Consistent with national data assessing the cumula-
tive onset of substance use among U.S. high schoolers (Monitoring the
Future, 2014), 55%, 64%, and 40% of our analytical sample had initiated
cigarette use, drunkenness, and marijuana use, respectively, by the 12th
grade. We focus on substance use initiation over other forms of offending for
two reasons. First, as is standard for self-report studies of delinquency,
PROSPER’s delinquency items have fixed reference periods (i.e., offending
in the past 12 months) and do not assess whether respondents had “ever”
committed various delinquent acts. A fixed reference period prevents an
accurate assessment of the timing of delinquency onset because it cannot
account for respondents who reported no delinquency at the first wave but
committed delinquency prior to the first wave’s 12-month reference period.
Second, substance use symbolizes autonomy and independence to youth
(Agnew, 2003; Moffitt, 1993), and it has been used in other studies that
focused on Moffitt’s theory of AL offending (Barnes & Beaver, 2010; Piquero
& Brezina, 2001).

Focal Independent Variables


To examine the influence of having LCP peers on respondents’ substance use
onset, we must first identify respondents who approximate this offending
group. Prior research has generally used two strategies to accomplish this.
The first strategy involves using a theory-based rules approach (Moffitt,
428 Crime & Delinquency 66(3)

Caspi, Dickson, Silva, & Stanton, 1996; Moffitt et al., 2002). Under this strat-
egy, researchers identify respondents who display either patterns of elevated
and sustained delinquency (e.g., Barnes, Beaver, & Boutwell, 2011), an early
onset of antisocial behavior (e.g., Simons, Wu, Conger, & Lorenz, 1994), or
the presence of serious or violent offending (e.g., Broidy et al., 2003) as a
proxy for LCP offenders. The second strategy involves using GBTM (Nagin,
2005) to detect meaningful subgroups of individuals who follow similar
delinquency trajectories over time (e.g., Blokland, Nagin, & Nieuwbeerta,
2005; Young, 2014). Although the theory-based and GBTM strategies have
unique strengths and weaknesses, studies comparing both strategies on the
same data show they generally produce similar results (Odgers et al., 2008;
Roisman et al., 2010).
Following the lead of other researchers who have used the PROSPER
study to examine Moffitt’s dual taxonomy (Rulison et al., 2014), we
employed a theory-based approach to identify persistently delinquent
youths, using a three-step process. First, we constructed measures of delin-
quency at each wave. During the in-school surveys, respondents reported
how many times in the past 12 months they had run away from home,
skipped school or class without an excuse, purposely damaged property,
avoided paying for things, stolen something from a store, stolen something
worth less than US$25, stolen something worth more than US$25, been
picked up by the police for breaking a law, broken into a building, thrown
rocks or bottles at someone, beaten up or physically fought someone, and
carried a hidden weapon. Response categories for these items were 1 =
never, 2 = once, 3 = twice, 4 = three or four times, and 5 = five or more
times (α = .81-.89). Item response theory (IRT) scaling was used to com-
bine the items (Osgood, McMorris, & Potenza, 2002); this ensures that the
scale is not dominated by less serious items.
Second, we identified a group of adolescents who displayed elevated and
sustained offending based on their IRT delinquency scores. This group, which
we call persistently delinquent, represents adolescents who displayed ele-
vated and sustained delinquency across Waves 1 to 8. These adolescents
scored at the 75th percentile or higher on the IRT delinquency scale at each
wave in which they had valid data (e.g., respondents who had five waves of
data scored at the 75th percentile at all five waves; respondents with six
waves of data scored at the 75th percentile at all six waves). According to this
definition, 3.25% of the sample fit this pattern of offending. Although our
group of persistently delinquent youth falls slightly short of the 5% to 8%
identified by Moffitt, other studies have estimated LCP groups of similar
sizes (Rulison et al., 2014; Young, 2014). Consistent with Moffitt’s LCP
group, our group of persistently delinquent youth scored high on a number of
Widdowson et al. 429

risk factors. Compared with the rest of our analytical sample, persistently
delinquent youth were more likely to be male (66% vs. 48% overall), reared
in single-parent households (33% vs. 19%), and free-school-lunch recipients
(38% vs. 27%). They also had poorer family relations (M = –.12 vs. 21),
lower parental discipline (M = 3.24 vs. 3.65), lower school grades (M = 3.89
vs. 4.31), and higher sensation seeking (M = 2.93 vs. 1.87).
Third, we linked each respondent to his or her friends and determined
whether he or she had a persistently delinquent friend. Persistently delin-
quent friend is a dichotomous variable indicating whether adolescents sent or
received a friendship nomination to or from a persistently delinquent peer at
each wave (0 = no, 1 = yes).5 In addition, we created measures indicating the
persistently delinquent friends’ own current substance use. Persistently delin-
quent friend uses cigarettes, persistently delinquent friend reports being
drunk, and persistently delinquent friend uses marijuana are dichotomous
variables indicating whether adolescents sent or received a friendship nomi-
nation from a persistently delinquent friend who also reported using the given
substance at that wave (0 = no, 1 = yes). Descriptively, over the course of
the entire study, 31% of respondents ever sent or received a friendship nomi-
nation to or from a persistently delinquent youth, and 16%, 14%, and 12% of
respondents ever had a persistently delinquent friend who used cigarettes,
reported being drunk, and used marijuana, respectively.

Control Variables
The control variables include respondents’ male gender (0 = female, 1 =
male), race/ethnicity (0 = Hispanic, African American, Asian, Native
American, or other non-White race, 1 = White), two-parent family (0 = other
family structure, 1 = two-parent family), and free or reduced price school
lunch (0 = no, 1 = yes). Time was captured as a set of dummy variables for
each grade level, with 6th grade serving as the reference category. The con-
trol variables also included variables capturing respondents’ school and fam-
ily background. School grades was a single item that asked respondents about
their grades in the past school year (1 = mostly lower than Ds to 5 = mostly
As). Positive family relations was a composite measure constructed by aver-
aging three standardized subscales that captured parent-child joint activities,
affective quality, and parental supervision (α = .82). Parental discipline was
the mean of five items measuring the extent to which respondents’ parents
utilized consistent parenting practices when the respondent misbehaved (α =
.75). Church attendance was a single item assessing how often respondents
attended church or religious services in the past year (1 = never to 4 = once
a week or more). Number of friends was the count of the number of incoming
430 Crime & Delinquency 66(3)

and outgoing friendship nominations made by respondents. Last, the controls


include two variables capturing individual differences in criminal propensity
(Gottfredson & Hirschi, 1990). Sensation seeking was the mean of three
items measuring respondents’ preference for risky and/or sensation seeking
experiences (α = .77). Delinquency was respondents’ IRT delinquency score
described above, which captures their involvement with a range of delinquent
acts. Table 1 provides descriptive statistics on the study variables. Except for
respondents’ gender and race/ethnicity (which are time-invariant), all predic-
tor variables in our analyses are time-varying and time-lagged by one wave
to establish the temporal order between peer influence and substance use
onset.

Analytical Strategy
We used discrete time event history models to predict substance use initia-
tion. Discrete time models treat time as an ordered categorical variable rather
than a continuous metric variable. This approach is well suited to data such
as ours where (a) the outcome is time to an event, (b) many cases are cen-
sored (in that the event never occurs), and (c) the measure of time consists of
a few relatively large intervals (in our case, yearly waves) versus many
smaller intervals (e.g., days, months; Allison, 1984). The models we esti-
mated took the following form:

 P 
ln  it  = α ( t ) + βX it −1 + βWi + βTimeit + βCommunityi + e,
 1 − Pit 

where Pit is the hazard (or conditional probability) that a respondent reports
using a particular substance at a wave, given that they never used that sub-
stance previously. The hazard is modeled as a function of βX it −1 , which is a
vector of time-varying predictors lagged by one wave to preserve temporal
order; βWi , which is a vector of time-invariant predictors (i.e., gender, race/
ethnicity); βTimeit , which represents a set of dummy variables for grade
level; and βCommunityi , which represents fixed effects for each community-
cohort to account for dependence. βs are interpreted as logistic coefficients.
This modeling strategy allowed us to test the duration-dependent associa-
tion between having a persistently delinquent friend and substance use initia-
tion while accounting for right-censoring and making no assumptions about
the underlying survival distribution (Allison, 1984). For these models, the
data file is structured in a person-wave (or person-grade) format. Respondents
were considered at risk for substance use onset starting at Wave 1 until they
either reported using a given substance or were still non-users at the time of
Widdowson et al. 431

Table 1. Descriptive Statistics on Study Variables.

M/% SE Minimum Maximum


Dependent variables
Cigarette use onset 0.072 0 1
Drunkenness onset 0.085 0 1
Marijuana use onset 0.057 0 1
Focal independent variables
PD friend 0.099 0 1
PD friend uses cigarettes 0.036 0 1
PD friend reports being drunk 0.031 0 1
PD friend uses marijuana 0.024 0 1
Control variables
Male 0.481 0 1
White 0.873 0 1
Two-parent family 0.797 0 1
Free or reduced price school lunch 0.235 0 1
School grades 4.121 0.004 1 5
Positive family relations −0.028 0.002 −2.354 1.001
Parental discipline 3.589 0.004 1 5
Church attendance 2.657 0.006 1 4
Number of friends 5.339 0.013 0 20
Sensation seeking 2.149 0.004 1 5
Delinquency 0.225 0.003 −0.341 3.943

Source. PROSPER Peers.


Note. Mean and percentages are the average values on the study variables across all
respondent-waves. N = 7,742 respondents with 58,950 respondent-waves. PD = persistently
delinquent; PROSPER = PROmoting School–Community–University Partnerships to Enhance
Resilience.

their final survey (i.e., were right-censored). About 5%, 1%, and 1% of
respondents had already smoked cigarettes, been drunk, and used marijuana,
respectively, at Wave 1 (i.e., were left-censored). Moffitt’s social mimicry
hypothesis pertains to the initiation of substance use (not maintenance or
escalation), so we excluded these respondents from their corresponding sub-
stance-specific onset analyses.

Results
Our first research question asks whether having a persistently delinquent
friend predicts the onset of substance use. Models 1 of Table 2 show a set of
discrete time models predicting cigarette use, drunkenness, and marijuana
Table 2. Discrete Time Coefficients Predicting Substance Use Initiation From PD Friend, PD Friend’s Use, and Controls.
Cigarette use onset Drunkenness onset Marijuana use onset

432
Model 1 Model 2 Model 1 Model 2 Model 1 Model 2

Predictors b SE b SE b SE b SE b SE b SE
PD friend 0.32*** 0.06 0.27*** 0.07 0.25*** 0.06 0.22*** 0.07 0.38*** 0.06 0.32*** 0.07
PD friend uses cigarettes 0.12 0.11
PD friend reports being drunk 0.11 0.10
PD friend uses marijuana 0.17 0.12
Male −0.42*** 0.04 −0.42*** 0.04 −0.30*** 0.03 −0.30*** 0.03 −0.15*** 0.04 −0.15*** 0.04
White 0.02 0.06 0.02 0.06 0.11* 0.06 0.11* 0.06 −0.09 0.06 −0.09 0.06
Two-parent family −0.13** 0.05 −0.13** 0.05 −0.13** 0.05 −0.13** 0.05 −0.15** 0.05 −0.15** 0.05
Free or reduced price school 0.13** 0.05 0.13** 0.05 −0.07 0.05 −0.07 0.05 −0.03 0.05 −0.03 0.05
lunch
School grades −0.29*** 0.02 −0.29*** 0.02 −0.14*** 0.02 −0.14*** 0.02 −0.19*** 0.03 −0.19*** 0.03
Positive family relations −0.26*** 0.05 −0.26*** 0.05 −0.27*** 0.05 −0.27*** 0.05 −0.31*** 0.05 −0.31*** 0.05
Parental discipline −0.09*** 0.02 −0.09*** 0.02 −0.08*** 0.02 −0.08*** 0.02 −0.08*** 0.02 −0.08*** 0.02
Church attendance −0.09*** 0.02 −0.09*** 0.02 −0.06*** 0.01 −0.06*** 0.01 −0.06*** 0.02 −0.06*** 0.02
Number of friends 0.01 0.01 0.01 0.01 0.06*** 0.01 0.06*** 0.01 0.01† 0.01 0.01† 0.01
Sensation seeking 0.22*** 0.02 0.22*** 0.02 0.27*** 0.02 0.27*** 0.02 0.23*** 0.02 0.23*** 0.02
Delinquency 0.43*** 0.03 0.43*** 0.03 0.46*** 0.03 0.46*** 0.03 0.57*** 0.03 0.57*** 0.03
7th grade (Wave 3) 0.49*** 0.07 0.49*** 0.07 0.83*** 0.10 0.83*** 0.10 0.97*** 0.13 0.97*** 0.13
8th grade (Wave 4) 0.80*** 0.07 0.80*** 0.07 1.61*** 0.09 1.61*** 0.09 1.58*** 0.12 1.57*** 0.12
9th grade (Wave 5) 0.83*** 0.07 0.82*** 0.07 2.30*** 0.09 2.30*** 0.09 1.98*** 0.12 1.97*** 0.12
10th grade (Wave 6) 0.85*** 0.08 0.85*** 0.08 2.44*** 0.09 2.43*** 0.09 2.16*** 0.12 2.14*** 0.12
11th grade (Wave 7) 1.00*** 0.08 1.00*** 0.08 2.49*** 0.09 2.49*** 0.09 2.26*** 0.12 2.25*** 0.12
12th grade (Wave 8) 1.09*** 0.08 1.09*** 0.08 2.57*** 0.10 2.57*** 0.10 2.35*** 0.12 2.34*** 0.12
Constant −1.60*** 0.18 −1.61*** 0.18 −3.62*** 0.18 −3.62*** 0.18 −3.48*** 0.22 −3.48*** 0.22
N respondents/N observations 7,347/39,080 7,347/39,080 7,648/39,935 7,648/39,935 7,688/44,394 7,688/44,394

Note. Models estimated on PROSPER Peers data. All models included fixed effects for community cohorts (untabled). PD = persistently delinquent; b = coefficient;
PROSPER = PROmoting School–Community–University Partnerships to Enhance Resilience.
†p < .10. *p < .05. **p < .01. ***p < .001 (two-tailed).
Widdowson et al. 433

use onset from persistently delinquent friend and the control variables. In line
with Moffitt’s (1993) social mimicry hypothesis, the results indicate that hav-
ing a persistently delinquent friend at one wave was significantly associated
with the risk of cigarette use initiation (b = 0.32, p < .001), drunkenness
initiation (b = 0.25, p < .001), and marijuana use initiation (b = 0.38, p <
.001) at the next wave, net of control variables. Specifically, adolescents who
had a friend who scored at the top 25% of delinquency at each wave were
between 28% and 46% more likely to subsequently initiate substance use
across 6th to 12th grades compared with adolescents without such friends, for
example, (exp(.38) – 1) × 100 = 46%.
Next, we tested our second research question, which asks whether persis-
tently delinquent friends’ own substance use predicts adolescents’ substance
use onset. As discussed, this research question assesses whether respondents’
modeling of persistently delinquent friends’ deviance is specific or general.
Models 2 of Table 2 show a series of discrete time models predicting cigarette
use, drunkenness, and marijuana use initiation from persistently delinquent
friend and persistently delinquent friends’ own substance use, net of controls.
These models assess whether friends’ use of specific substances predicts the
onset of use of those same substances above and beyond the impact of simply
having a persistently delinquent friend. Because all respondents who had a
persistently delinquent friend who smoked cigarettes, for example, also by
definition had a persistently delinquent friend, the former coefficient (when
both variables are in the model) represents the effect of having a persistently
delinquent friend who smokes relative to a persistently delinquent friend who
does not smoke, and the latter coefficient represents the effect of having a
persistently delinquent friend who does not smoke relative to not having a
persistently delinquent friend. Thus, the models indicate whether having a
persistently delinquent friend is associated with cigarette use initiation
because that friend smokes.
The results from Models 2 show that a persistently delinquent friend’s
own cigarette use (b = 0.12, p > .10), drunkenness (b = 0.11, p > .10), and
marijuana use (b = 0.17, p > .10) were positively but not significantly asso-
ciated with adolescents’ subsequent initiation of those substances, indicating
that there is no difference between having a persistently delinquent friend
who uses a substance versus having a persistently delinquent friend who does
not in predicting adolescent substance use onset.7 To further illustrate this
finding, Figure 1 shows the predicted probability of cigarette use, drunken-
ness, and marijuana use onset from 6th to 12th grades for those who did not
have a persistently delinquent friend, those who had a persistently delinquent
friend who did not use substances, and those who had a persistently delin-
quent friend who used substances, with all control variables held constant at
their means. The figure illustrates that having a persistently delinquent friend
434 Crime & Delinquency 66(3)

Figure 1. Predicted probability of substance use initiation from 6th to 12th grades
by PD friendship status in the PROSPER study.
Note. PD = persistently delinquent; PROSPER = PROmoting School–Community–University
Partnerships to Enhance Resilience.
Widdowson et al. 435

increases the risk of substance use onset, but there is not a significant addi-
tional difference depending on whether the friend used that substance or not.
Together, these findings indicate that adolescents with a persistently delin-
quent friend tended to initiate substance use regardless of whether the same
friend used that particular substance or not.

Sensitivity Analyses
Our main analyses used a theory-based approach to identify respondents who
approximate Moffitt’s description of LCP offenders. However, a criticism of
this approach is that the criteria and cutoffs used to identify offending groups
are inherently subjective. Thus, it is important to assess whether our results
are sensitive to the method used to identify LCP offenders. To this end, we
conducted two sensitivity tests. First, we used a different criterion that identi-
fied respondents who displayed elevated and sustained violence (instead of
delinquency more generally), given that Moffitt suggests violent offending is
a key feature of LCP offending. In particular, we identified respondents who
reported engaging in at least one violent act at each wave (i.e., thrown rocks
or bottles at someone, beaten up or physically fought someone, and carried a
hidden weapon). According to this definition, 3.36% of the sample was
defined as persistently violent. We then linked respondents to their friends’
status and reestimated our models using this different theory-based approach.
Panel A of Table 3 provides a summary of the findings. The first model pre-
dicts substance use initiation from the persistently violent friend, and the sec-
ond model includes persistently violent friends’ own use. Altogether, six
models are shown in Panel A, each of which includes the full set of control
variables and fixed effects for community-cohort. Similar to the main analy-
ses, Table 3 indicates that having a persistently violent friend was associated
with a higher risk of substance use, and adolescents tended to initiate sub-
stance use regardless of whether their persistently violent friend used that
particular substance or not. Second, we tested whether our main findings
were sensitive to the use of the 75th percentile to define elevated offending
by creating separate measures of persistent delinquency using the 70th and
80th percentiles. We then linked respondents to their friends’ status and rees-
timated our models using these different theory-based cutoffs. Panels B and
C of Table 3 provide a summary of the findings. Altogether, 12 models are
shown, each of which includes the full set of control variables and commu-
nity-cohort fixed effects. In both cases, the results were substantively similar
to the main findings using the 75th percentile. Together, the results of these
sensitivity analyses are consistent with the conclusions from the main
analysis.
436 Crime & Delinquency 66(3)

Table 3. Supplementary Discrete Time Coefficients Predicting Substance Use


Initiation From Alternative Measures of PD Friend and PD Friend’s Use.

Model 1 Model 2

b SE b SE

Panel A: PV friend defined as reporting at least one violent act at every wave
Cigarette use initiation
  PV friend 0.40*** 0.06 0.38*** 0.07
   PV friend uses cigarettes 0.07 0.11
Drunkenness initiation
  PV friend 0.31*** 0.06 0.29*** 0.07
   PV friend reports being drunk 0.04 0.10
Marijuana use initiation
  PV friend 0.42*** 0.06 0.38*** 0.07
   PV friend uses marijuana 0.15 0.12
Panel B: PD friend defined as scoring at the 70th percentile at every wave
Cigarette use initiation
  PD friend 0.28*** 0.05 0.24*** 0.06
   PD friend uses cigarettes 0.09 0.10
Drunkenness initiation
  PD friend 0.24*** 0.05 0.22*** 0.06
   PD friend reports being drunk 0.10 0.09
Marijuana use initiation
  PD friend 0.34*** 0.05 0.30*** 0.06
   PD friend uses marijuana 0.13 0.11
Panel C: PD friend defined as scoring at the 80th percentile at every wave
Cigarette use initiation
  PD friend 0.31*** 0.07 0.31*** 0.10
   PD friend uses cigarettes 0.02 0.15
Drunkenness initiation
  PD friend 0.27*** 0.07 0.21* 0.09
   PD friend reports being drunk 0.09 0.13
Marijuana use initiation
  PD friend 0.35*** 0.08 0.28** 0.09
   PD friend uses marijuana 0.14 0.16

Source. PROSPER Peers.


Note. The table displays the results of 18 separate models. Each model includes the full
set of control variables and community-cohort fixed effects (untabled). b = coefficient;
PV = persistently violent; PD = persistently delinquent; PROSPER = PROmoting School–
Community–University Partnerships to Enhance Resilience.
†p < .10. *p < .05. **p < .01. ***p < .001 (two-tailed).
Widdowson et al. 437

Discussion
Past research evaluating Moffitt’s (1993) dual taxonomy has focused almost
entirely on the identification of different life-course trajectories and the etiol-
ogy of the different groups (see Moffitt, 2006; Piquero et al., 2013). However,
no study has directly examined one of the theory’s central hypotheses con-
cerning the etiology of AL delinquency: the social mimicry hypothesis.
According to Moffitt, the sudden rise in offending observed in adolescence is
the result of AL delinquents experiencing a “maturity gap” and mimicking
the behavior of their LCP peers in an attempt to appear more adult-like and
independent, and this mimicking, or modeling, can occur in the context of
adolescents’ social networks.
The goal of this study was to test two related research questions. The first
addressed whether having a persistently delinquent friend was associated
with a higher risk of substance use initiation from 6th to 12th grades. Across
three forms of substance use, we found that having a persistently delinquent
friend was associated with a higher risk of onset. The second addressed
whether social mimicry was general or particular to the friends’ use of spe-
cific substances. Our findings indicate that adolescents who have a persis-
tently delinquent friend are just as likely to initiate substance use whether the
persistently delinquent friend uses the same substance or not.
The results from these two research questions can be taken as providing
support for the social mimicry hypothesis, but the support is limited. Our
study does suggest that having a persistently delinquent friend is associated
with an increased hazard of substance use initiation. At the same time, how-
ever, it appears that adolescents tend to initiate substance use regardless of
their persistently delinquent friends’ own use. Although this pattern of find-
ings does not refute the social mimicry hypothesis, it does call into question
the premise behind it and the causal mechanism through which mimicry
occurs (i.e., AL youth should initiate the specific behaviors that are displayed
by their persistently delinquent friends). Such a finding, then, raises the ques-
tion: Why is having a delinquent friend a risk factor for substance use onset
if not because of the friends’ own substance use? One possibility is that ado-
lescents might attempt to demonstrate “maturity” to their friends by using
substances even though their persistently delinquent friend does not currently
use that particular substance. Adolescents, for example, might see their delin-
quent friends shoplift and assume that friend engages in analogous behaviors,
such as substance use. If this assumption, in turn, influences substance use
initiation, it may be one way a non-using persistently delinquent friend influ-
ences individuals’ own use. Such a possibility would be consistent with stud-
ies that show that measures of delinquent peer association that do not contain
438 Crime & Delinquency 66(3)

peer substance use still predict individual substance use (e.g., Dishion,
Capaldi, Spracklen, & Li, 1995; Vitaro, Brendgen, Ladouceur, & Tremblay,
2001) and with work that views delinquent peer groups as the impetus for
general rule-breaking behavior (Dishion, 2000). Future work should further
unpack the mechanisms by which adolescents influence one another, espe-
cially with respect to deviant behavior.
The PROSPER study did not contain measures that could accurately
assess the onset of delinquent behavior. Thus, we were unable to examine
delinquency initiation. However, because the social mimicry hypothesis
would make similar predictions regarding both delinquency and substance
use initiation, we expect that our findings would generalize to delinquent
outcomes as well. We encourage researchers to use other longitudinal peer
network data sources that capture the onset of delinquent behavior to study
this component of the social mimicry hypothesis. We note that many forms of
delinquency appear earlier in the life course than substance use (Elliott, 1994;
Loeber et al., 2012). Thus, researchers may need to study forms of delin-
quency that have their onset in the mid- and late-teen years, such as serious
forms of violence and economic crime (Loeber et al., 2012). Such examina-
tions may be particularly important in light of other elements of Moffitt’s
(1993) theory, such as the maturity gap, which should be most pronounced—
and thus most criminogenic—during adolescence.
Researchers studying peer influence have sometimes suggested that tar-
geting the instigator (or “ringleader”) of criminal mischief provides the best
result for reducing overall crime. According to Moffitt’s (1993) theory, this
would consist of targeting offenders who display elevated and sustained
forms of offending (i.e., LCP offenders) whom the majority of youthful
offenders are thought to mimic. Although our finding of limited mimicry
effects suggests that peer influence from persistently delinquent youth might
be more complicated than originally spelled out by Moffitt, the fact that per-
sistently delinquent youth influenced respondents’ substance use initiation
regardless of their own use suggests that policies targeting such offenders
may be a fruitful way to reduce offending. That is, interventions targeting
persistently delinquent adolescents may have cascading benefits that go
beyond simply reducing a persistently delinquent’s own deviant behavior.
The influence of persistently delinquent youth on AL offenders also illus-
trates the relevance of collecting information on friends’ histories when con-
ducting risk assessments for delinquency initiation.
Although our findings advance the literature, we caution that our study
only provides a partial test of Moffitt’s (1993) social mimicry hypothesis.
Several design features could have made our study a conservative test, which
may have failed to detect evidence in favor of the social mimicry hypothesis.
Widdowson et al. 439

First, as noted, it is possible that social mimicry happens from afar and
requires no friendship ties. In such a scenario, adolescents might observe an
LCP offender using substances at a party or after school despite not being
friends, or they might hear about such use through third-party communica-
tion. Second, our findings do not rule out the possibility that respondents’
nonpersistently delinquent friends were first influenced by persistently delin-
quent youth and then, in turn, influenced respondents. Such a possibility
would imply a more complicated social network diffusion model of social
mimicry (Rulison, Gest, & Osgood, 2015). In such a scenario, a small num-
ber of LCP youth would influence a small number of AL youth, who, in turn,
would influence a growing number of other AL youth. Third, our findings do
not rule out the possibility of persistently delinquent youth influence from
other grade levels. Like other social network studies (e.g., NSCR School
Project), the PROSPER data are constrained to same-grade friendship ties.
Prior research has emphasized the role of older and more experienced offend-
ers who recruit younger offenders (Reiss & Farrington, 1991). Fourth, it is
possible that persistently delinquent friends might influence other forms of
substance use, especially hard drugs. Hagan (1991) argued that “party” sub-
culture characterized by drinking and smoking is relatively normal during
adolescence, whereas more serious offending (which might include heavy
drug use) is not. If any of these four scenarios is true, it could explain why our
findings did not strongly support the social mimicry hypothesis. We consider
these possibilities important avenues for future researchers considering the
social mimicry hypothesis.
Although our study contributes to the literature, it has limitations. First,
our study assessed friends’ delinquency and substance use via friends’ own
reports of their behavior. This technique avoids potential problems such as
projection bias, or the tendency for a respondent’s reports about their friends
to be influenced by the respondent’s own behavior (Prinstein & Wang, 2005).
Yet, one disadvantage of the method is that we could not directly determine
whether respondents observed or knew about their friends’ delinquency and
substance use. Future work on this topic should assess the extent of youths’
knowledge and observation of the friend behaviors that theoretically are
being mimicked. Second, the identification of delinquency groups in our
study is limited to a 5- to 8-year period. Although we tested the validity of the
groups and conducted sensitivity tests using different criteria and different
thresholds for assigning members of the sample to the persistently delinquent
group, there is still some controversy over how to best define LCP status
(Piquero, 2008). Therefore, it is possible that our findings might differ if we
were able to observe the sample for longer periods of time. We consider this
an area of future research. Third, our data come from a largely rural sample.
440 Crime & Delinquency 66(3)

As some have pointed out, such samples might lack large numbers of fre-
quent and serious offenders, which has implications for the generalizability
of the findings (Osgood & Chambers, 2000). We suspect that our sample is
not problematic, however, given a recent meta-analysis of peer influence and
selection for delinquency by Gallupe, McLevey, and Brown (2019), in which
estimates for our sample (Osgood, Feinberg, & Ragan, 2015) were excep-
tionally close to the composite results across diverse samples. Even so, future
research should consider replicating our findings in other samples and con-
texts. Fourth, while the use of the PROSPER data is advantageous for our
study due to longitudinal network data, PROSPER lacks any measures of the
maturity gap. Thus, we test only the social mimicry hypothesis as opposed to
the entirety of the theoretical propositions regarding AL offending. Future
research should consider both aspects together.
In conclusion, our study contributes to the literature by providing an empiri-
cal test of Moffitt’s (1993) social mimicry hypothesis. We interpret our results
as demonstrating limited support for the hypothesis. The fact that adolescents
who had a persistently delinquent friend tended to initiate substance use regard-
less of whether the persistently delinquent friend used that specific substance
undermines the contention that AL delinquents directly mimic their LCP
friends’ behavior. Whether our findings would replicate with different samples,
measures, and contexts is a matter that future researchers should consider.

Authors’ Note
An earlier draft of this article was presented at the 2015 American Society of
Criminology Meeting in Washington, D.C.

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: Grants from the W. T. Grant Foundation
(8316), National Institute on Drug Abuse (R01-DA018225), and National Institute of
Child Health and Development (R24-HD041025) supported this research. The analy-
ses used data from PROmoting School–Community–University Partnerships to
Enhance Resilience (PROSPER), a project directed by R. L. Spoth, funded by the
National Institute on Drug Abuse and the National Institute on Alcohol Abuse and
Alcoholism (AA14702). The content is solely the responsibility of the authors and
does not necessarily represent the official views of the National Institutes of Health.
Widdowson et al. 441

ORCID iD
Alex O. Widdowson https://orcid.org/0000-0002-1501-7986

Notes
1. The PROmoting School–Community–University Partnerships to Enhance
Resilience (PROSPER) sample was part of a place-randomized substance abuse
prevention trial. School districts assigned to the treatment condition received
additional family- and school-based substance use programming. To determine
whether our results were sensitive to this design feature, we conducted supple-
mentary analyses that tested (a) whether the effects were moderated by the
treatment condition and (b) whether the results were affected by controlling the
treatment condition in the main analysis. We found that the results were unaf-
fected by both.
2. As crime rates are generally lower in rural areas, our sample may lack a high
number of life-course persistent (LCP) peers.
3. Given that the PROSPER social networks were restricted to same school and
grade friendship ties, our findings do not capture influence from friends outside
of school or older/younger peers.
4. For these supplemental analyses, we retained those who were present for at least
four out of five waves during 6th to 9th grades. Persistently delinquent youth
were then defined as scoring at the 75th percentile (or higher) on delinquency at
each wave they were present.
5. We focus on send-and-receive network ties to examine peer delinquency, but the
results were consistent with the conclusions from the main analysis when we
used only send-network ties.
6. Our measure of persistently delinquent friend does not capture possible mimicry
“from afar” (Moffitt, 1993: 688).
7. We estimated three additional models that predicted cigarette, drunkenness, and
marijuana onset from persistently delinquent friend and a new global measure
of whether respondents’ persistently delinquent friend either used cigarettes,
reported being drunk, or used marijuana. We did this to assess the possibility
that adolescents are influenced by their persistently delinquent friend’s “broader
use” of substances (e.g., marijuana use by a persistently delinquent friend might
influence drunkenness or cigarette use onset by an adolescence-limited [AL]
delinquent). Those models showed that persistently delinquent friends’ global
substance use did not predict substance use onset.

References
Agnew, R. (2003). An integrated theory of the adolescent peak in offending. Youth &
Society, 34, 263-299.
Akers, R. L., & Sellers, C. S. (2004). Criminological theories: Introduction, evalua-
tion, and application (4th ed.). Los Angeles, CA: Roxbury.
Allison, P. D. (1984). Event history analysis. Beverly Hills, CA: Sage.
442 Crime & Delinquency 66(3)

Bachman, J. G., O’Malley, P. M., Schulenberg, J. E., Johnston, L. D., Bryant, A. L., &
Merline, A. C. (2002). The decline of substance use in young adulthood: Changes
in social activities, roles, and beliefs. London, England: Psychology Press.
Barnes, J. C., & Beaver, K. M. (2010). An empirical examination of adolescent-lim-
ited offending: A direct test of Moffitt’s maturity gap thesis. Journal of Criminal
Justice, 38, 1176-1185.
Barnes, J. C., Beaver, K. M., & Boutwell, B. B. (2011). Examining the genetic under-
pinnings to Moffitt’s developmental taxonomy: A behavioral genetic analysis.
Criminology, 49, 923-954.
Bellair, P. E., McNulty, T. L., & Piquero, A. R. (2016). Verbal ability and persistent
offending: A race-specific test of Moffitt’s theory. Justice Quarterly, 33, 455-
480.
Blokland, A. A., Nagin, D., & Nieuwbeerta, P. (2005). Life span offending trajecto-
ries of a Dutch conviction cohort. Criminology, 43, 919-954.
Broidy, L. M., Nagin, D. S., Tremblay, R. E., Bates, J. E., Brame, B., Dodge, K.
A., . . . Lynam, D. R. (2003). Developmental trajectories of childhood disrup-
tive behaviors and adolescent delinquency: A six-site, cross-national study.
Developmental Psychology, 39, 222-245.
Dijkstra, J. K., Kretschmer, T., Pattiselanno, K., Franken, A., Harakeh, Z., Vollebergh,
W., & Veenstra, R. (2015). Explaining adolescents’ delinquency and substance
use: A test of the maturity gap: The SNARE study. Journal of Research in Crime
and Delinquency, 52, 747-767.
Dinges, M. M., & Oetting, E. R. (1993). Similarity in drug use patterns between ado-
lescents and their friends. Adolescence, 28, 253-266.
Dishion, T. J. (2000). Cross-setting consistency in early adolescent psychopathology:
Deviant friendships and problem behavior sequelae. Journal of Personality, 68,
1109-1126.
Dishion, T. J., Capaldi, D., Spracklen, K. M., & Li, F. (1995). Peer ecology of male
adolescent drug use. Development and Psychopathology, 7, 803-824.
Elliott, D. S. (1994). Serious violent offenders: Onset, developmental course, and
termination—The American Society of Criminology 1993 presidential address.
Criminology, 32, 1-21.
Franken, A., Harakeh, Z., Veenstra, R., Vollebergh, W., & Dijkstra, J. K. (2017).
Social status of adolescents with an early onset of externalizing behavior: The
SNARE study. The Journal of Early Adolescence, 37, 1037-1053.
Gallupe, O., McLevey, J., & Brown, S. (2019). Selection and influence: A meta-
analysis of the association between peer and personal offending. Journal of
Quantitative Criminology, 35, 313-335.
Gottfredson, M. R., & Hirschi, T. (1990). A general theory of crime. Palo Alto, CA:
Stanford University Press.
Hagan, J. (1991). Destiny and drift: Subcultural preferences, status attainments, and
the risks and rewards of youth. American Sociological Review, 56, 567-582.
Haynie, D. L. (2001). Delinquent peers revisited: Does network structure matter?
American Journal of Sociology, 106, 1013-1057.
Widdowson et al. 443

Huba, G. J., Wingard, J. A., & Bentler, P. M. (1979). Beginning adolescent drug
use and peer and adult interaction patterns. Journal of Consulting and Clinical
Psychology, 47, 265-276.
Jennings, W. G., & Reingle, J. M. (2012). On the number and shape of developmen-
tal/life-course violence, aggression, and delinquency trajectories: A state-of-the-
art review. Journal of Criminal Justice, 40, 472-489.
Loeber, R., Menting, B., Lynam, D. R., Moffitt, T. E., Stouthamer-Loeber, M.,
Stallings, R., . . .Pardini, D. (2012). Findings from the Pittsburgh Youth Study:
Cognitive impulsivity and intelligence as predictors of the age–crime curve.
Journal of the American Academy of Child & Adolescent Psychiatry, 51,
1136-1149.
Moffitt, T. E. (1993). Adolescence-limited and life-course persistent antisocial behav-
ior: A developmental taxonomy. Psychological Review, 100, 647-701.
Moffitt, T. E. (2006). A review of research on the taxonomy of life-course persistent
versus adolescence-limited antisocial behavior. In F. T. Cullen, J. Wright & K.
Blevins (Eds.), Taking stock: The status of criminological theory (pp. 277-312).
Piscataway, NJ: Transaction Publishers.
Moffitt, T. E., Caspi, A., Dickson, N., Silva, P. A., & Stanton, W. (1996). Childhood-
onset versus adolescent-onset antisocial conduct problems in males: Natural his-
tory from ages 3 to 18 years. Development and Psychopathology, 8, 399-424.
Moffitt, T. E., Caspi, A., Harrington, H., & Milne, B. J. (2002). Males on the life-
course-persistent and adolescence-limited antisocial pathways: Follow-up at age
26 years. Development and Psychopathology, 14, 179-207.
Moffitt, T. E., Lynam, D. R., & Silva, P. A. (1994). Neuropsychological tests predict-
ing persistent male delinquency. Criminology, 32, 277-300.
Monitoring the Future. (2014). Trends in lifetime prevalence of use of various drugs
in grades 8, 10, and 12. Retrieved from http://www.monitoringthefuture.org/
data/14data/14drtbl1.pdf
Nagin, D. (2005). Group-based modeling of development. Cambridge, MA: Harvard
University Press.
Odgers, C. L., Moffitt, T. E., Broadbent, J. M., Dickson, N., Hancox, R. J., Harrington,
H., . . . Caspi, A. (2008). Female and male antisocial trajectories: From child-
hood origins to adult outcomes. Development and Psychopathology, 20, 673-716.
Osgood, D. W., & Chambers, J. M. (2000). Social disorganization outside the metrop-
olis: An analysis of rural youth violence. Criminology, 38, 81-116.
Osgood, D. W., Feinberg, M. E., & Ragan, D. T. (2015). Social networks and the dif-
fusion of adolescent problem behavior: Reliable estimates of selection and influ-
ence from 6th through 9th grade. Prevention Science, 16, 832-843.
Osgood, D. W., McMorris, B. J., & Potenza, M. T. (2002). Analyzing multiple-item
measures of crime and deviance I: Item response theory scaling. Journal of
Quantitative Criminology, 18, 267-296.
Petraitis, J., Flay, B. R., & Miller, T. Q. (1995). Reviewing theories of adolescent sub-
stance use: Organizing pieces in the puzzle. Psychological Bulletin, 117, 67-86.
444 Crime & Delinquency 66(3)

Piquero, A. R. (2008). Taking stock of developmental trajectories of criminal activity


over the life course. In A. M. Liberman (Ed.), The long view of crime: A synthesis
of longitudinal research (pp. 23-78). New York, NY: Springer.
Piquero, A. R., & Brezina, T. (2001). Testing Moffitt’s account of adolescence-lim-
ited delinquency. Criminology, 39, 355-370.
Piquero, A. R., Diamond, B., Jennings, W. G., & Reingle, J. M. (2013). Adolescence-
limited offending. In C. L. Gibson & M. D. Krohn (Eds.), Handbook of life-
course criminology (pp. 129-142). New York, NY: Springer.
Prinstein, M. J., & Wang, S. S. (2005). False consensus and adolescent peer con-
tagion: Examining discrepancies between perceptions and actual reported lev-
els of friends’ deviant and health risk behaviors. Journal of Abnormal Child
Psychology, 33, 293-306.
Raine, A., Moffitt, T. E., Caspi, A., Loeber, R., Stouthamer-Loeber, M., & Lynam, D.
(2005). Neurocognitive impairments in boys on the life-course persistent antiso-
cial path. Journal of Abnormal Psychology, 114, 38-49.
Reiss, A. J., Jr., & Farrington, D. P. (1991). Advancing knowledge about co-offend-
ing: Results from a prospective longitudinal survey of London males. Journal of
Criminal Law and Criminology, 82, 360-395.
Roisman, G. I., Monahan, K. C., Campbell, S. B., Steinberg, L., & Cauffman, E., &
The National Institute of Child Health Human Development Early Child Care
Research Network. (2010). Is adolescence-onset antisocial behavior develop-
mentally normative? Development and Psychopathology, 22, 295-311.
Rubin, D. B. (1987). Multiple imputation for nonresponse in surveys. New York, NY:
John Wiley.
Rulison, K. L., Kreager, D. A., & Osgood, D. W. (2014). Delinquency and peer
acceptance in adolescence: A within-person test of Moffitt’s hypotheses.
Developmental Psychology, 50, 2437-2448.
Rulison, K. L., Gest, S. D., & Osgood, D. W. (2015). Adolescent peer networks and
the potential for the diffusion of intervention effects. Prevention Science, 16,
133-144.
Simons, R. L., Wu, C. I., Conger, R. D., & Lorenz, F. O. (1994). Two routes to delin-
quency: Differences between early and late starters in the impact of parenting and
deviant peers. Criminology, 32, 247-276.
Spoth, R., Greenberg, M., Bierman, K., & Redmond, C. (2004). PROSPER com-
munity–university partnership model for public education systems: Capacity-
building for evidence-based, competence-building prevention. Prevention
Science, 5, 31-39.
Spoth, R., Redmond, C., Shin, C., Greenberg, M., Clair, S., & Feinberg, M. (2007).
Substance use outcomes at 18 months past baseline: The PROSPER Community-
University Partnership Trial. American Journal of Preventive Medicine, 32,
395-402.
Tibbetts, S. G., & Piquero, A. R. (1999). The influence of gender, low birth weight,
and disadvantaged environment in predicting early onset of offending: A test of
Moffitt’s interactional hypothesis. Criminology, 37, 843-878.
Widdowson et al. 445

Vitaro, F., Brendgen, M., Ladouceur, R., & Tremblay, R. E. (2001). Gambling, delin-
quency, and drug use during adolescence: Mutual influences and common risk
factors. Journal of Gambling Studies, 17, 171-190.
Warr, M. (1996). Organization and instigation in delinquent groups. Criminology,
34, 11-37.
Weerman, F. M., & Smeenk, W. H. (2005). Peer similarity in delinquency for different
types of friends: A comparison using two measurement methods. Criminology,
43, 499-524.
Young, J. T. (2014). “Role magnets”? An empirical investigation of popularity trajec-
tories for life-course persistent individuals during adolescence. Journal of Youth
and Adolescence, 43, 104-115.

Author Biographies
Alex O. Widdowson is an assistant professor of Criminal Justice at the University of
Louisville. His research interests include the development of crime over the life
course, the consequences of criminal behavior and criminal justice sanctioning, peer
delinquency, and prisoner reentry.
J. W. Andrew Ranson is an assistant professor in the Department of Social and
Cultural Sciences at Marquette University. His research primarily focuses on varia-
tion in crime and social control across communities.
Sonja E. Siennick, PhD, 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.
Kelly L. Rulison is an associate professor in the Department of Public Health
Education at the University of North Carolina Greensboro. Her research interests cen-
ters around three themes: identifying how social networks shape the development of
health risk behaviors, identifying when and how peer networks facilitate the diffusion
of intervention effects, and optimizing interventions so that they have a greater public
health impact.
D. Wayne Osgood is Professor Emeritus of Criminology and Sociology at
Pennsylvania State University. His research interests have included peers and delin-
quency, time use and offending, crime and the life course, evaluating programs to
prevent and reduce delinquency, and quantitative research methods.

You might also like