Abstract

IntroductionAdolescent substance use and delinquency are major public health problems.
Although communitybased prevention strategies have been recommended to produce population
level reductions in rates of substance use and delinquency, few models show evidence of effectiveness.

PurposeTo test the efficacy of a communitybased prevention system, Communities That Care
(CTC), in reducing community rates of problem behaviors, particularly effects on specific profiles
of adolescent substance use and delinquency in eighth and tenth graders.

MethodsTwenty-four communities were randomized to CTC intervention or control groups. Data

were collected from 14,099 eighth and tenth grade students in these communities using anonymous
cross-sectional surveys in 2004 and 2010 and analyzed in 2012. Outcomes were four different
profiles of selfreported substance use and delinquency in eighth grade and five profiles in tenth
grade.

ResultsIn the cross-sectional 2010 data, there was no intervention effect on the probability of
experimenting with substances or of substance use coupled with delinquent activities for either grade.
However, tenth graders in intervention communities were significantly less likely to be alcohol users
than those in control communities (OR=0.69, CI=0.48, 1.00).

ConclusionsCrosssectional population surveys showed evidence of CTC effects in reducing tenth

grade alcohol users but not experimenters. A community-wide reduction in adolescent alcohol use is
important because alcohol is the most commonly used illicit substance during adolescence, and early
initiation of alcohol use has been associated with alcohol-related disorders in adulthood. Failure to find
hypothesized effects on experimenters qualifies these results.

Introduction
Adolescence is marked by increasing involvement in risk-seeking behaviors such as
1,2
substance use and delinquency. Such behaviors can result in negative consequences such
as poor academic performance, compromised health, and increased rates of victimization and
3,4
physical injury. Initiation of alcohol use during adolescence increases risk for alcohol and
5,6
other drug disorders in adulthood, and early initiation of delinquency predicts continued
7,8
involvement in severe, violent offending. The high financial costs of these problem
behaviors are related to medical care, work loss, drug treatment programs, and correctional
911
systems.
Many interventions have been developed and demonstrated effective in preventing
12
adolescent problem behaviors. However, relatively few communitybased preventive
interventions have shown effectiveness, however, leading federal agencies to call for more
10,12,14,15
development and evaluation of such approaches. Compared with single
interventions, community-based strategies have the potential to affect multiple outcomes
across a larger population, given their focus on simultaneously addressing numerous risk
and protective factors via the implementation of multiple, locally coordinated preventive
16,17
interventions that are likely to affect the majority of residents in a community.
To date, only two communitybased prevention systems have been found to impact
18
adolescent delinquency and/or substance use. The first is the Promoting School
CommunityUniversity Partnerships to Enhance Resilience (PROSPER) project, which relies
on university staff, school district officials, and other community members to implement
familyfocused and schoolbased prevention services. This model has demonstrated
19
reductions in the initiation, prevalence, and escalation of substance use during adolescence.
20,21
The second, Communities That Care (CTC), relies on local coalitions to implement a
range of services that reduce risk factors and strengthen protective factors experienced by
youth. CTC has been shown to reduce the initiation of tobacco use, alcohol use, delinquency,
22,23
and violence among a longitudinal panel of students followed from grade five to ten.
Previous tests of communitybased prevention systems examine average effects on each
behavioral outcome. However, etiologic evidence indicates that risk and protective factors
24,25
predict numerous problem behaviors that often co-occur. Thus, when evaluating
intervention effects, it would be useful to examine the degree to which an intervention
affects involvement in multiple problem behaviors, such as substance use and violence, as
well as combinations of such behaviors. The current study investigates the degree to which
CTC affects the probability that adolescents will engage in specific behavioral profiles of
substance use, delinquency, and violence.
Although the methodology described in this paper could be applied to the evaluation of any
type of preventive intervention, it is particularly relevant for comprehensive approaches that
involve the implementation of multiple services targeting diverse predictors of related
outcomes. In the CTC system, community coalitions receive training and technical assistance
and follow structured protocols to prevent youth problem behaviors. Based on a public health
framework, CTC assists communities in conducting population surveys of public secondary
school students to identify locally elevated risk factors and suppressed protective factors and
in implementing tested and effective preventive interventions that target these factors. Thus,
the interventions are specific to the needs of each community and include one or more types
of services such as school-based curricula and organizational change models, afterschool
services (e.g., tutoring and mentoring programs), and family-focused interventions such as
parenting workshops.

Coalitions are encouraged to deliver an array of preventive interventions, which target to

change multiple risk and protective factors in as many participants as possible, so long as the
interventions are implemented with fidelity and in a coordinated fashion to minimize
duplication of services. Although communities may implement interventions targeting higher-
risk youth (i.e., selective or indicated prevention programs), CTC emphasizes the delivery of
services to the general population. Therefore, the CTC system can be considered a universal
prevention approach intended to reduce communitywide problem behaviors.

The CTC theory of change posits that this system will produce reductions in the overall
prevalence of health problems with similar effects across individuals. However, the effects
of CTC could be specific to certain groups. For example, evaluations of other universal
interventions have reported stronger or weaker impacts on problem behaviors according to
2632
demographic status (e.g., sex or race/ethnicity) or for higher- versus lower-risk youth.
The current study examines outcomes in a different manner. Rather than assessing the
degree to which the intervention has produced changes in a single behavior for groups of
individuals defined according to demographic characteristics or risk status, we investigate
 the degree to which the CTC system affects the probability that adolescents engage in
specific behavioral profiles of substance use, delinquency, and violence.

This approach is based on prior examination of cross-sectional data from the CTC study, which
33
identified exclusive profiles of multiple problem behaviors. Students in the abstainer profile had
very low probabilities of reporting any substance use, delinquency, or violence; experimenters
reported some lifetime substance use but little recent use, and were very unlikely to have engaged
in delinquent behaviors; students in the drug use profile were very likely to report current drug use
but less likely to engage in delinquent behaviors; and problem students were very likely to report
substance use and delinquent activities. There was also evidence for a fifth profile: a group of
students characterized as alcohol users, who were very likely to engage in current alcohol use but
had limited involvement with other substances or delinquency, and primarily came from the
experimenter profile when only four profiles were estimated. Evidence that students engage in
different combinations of problem behaviors provides the foundation for this more nuanced
evaluation of how CTC can effect changes in communities. Investigating the degree to which CTC
or any intervention impacts profiles of co-occurring behaviors can better illuminate the ways in
which the intervention achieves change and better specify where the strongest intervention effects
occur, although the methods are limited by the assumption that the profiles are stable across
samples and across time. This approach jointly tests intervention effects on the probability
that study participants are in each of the identified categories while also allowing tests of
specific hypotheses about the impacted groups.

Research Aims
Although effects of CTC have been shown in a longitudinal panel, they have not been
demonstrated with cross-sectional data and groups of students most impacted by the
intervention have not been identified. The current paper uses repeated crosssectional data to
examine the probability of eighth and tenth grade students in intervention and control
communities engaging in different profiles of problem behaviors. Prior analyses have shown
that in 2004, before implementation of preventive interventions selected through the CTC
33
process, there was no difference in profiles among eighth grade students. This paper shows
evidence for the comparability of these profiles in 2004 surveys of tenth graders. The primary
aim of this study was to examine effects of the CTC intervention on cross-sectional profiles
of problem behaviors in 2010, 6 years after the intervention communities began targeted
prevention services. Given that CTC generally advocates for implementation of universal
preventive interventions, we previously hypothesized that the greatest preventive impact
33
would be localized to youth engaging in more common and less serious problem behaviors.
We test the hypothesis that in contrast to students in control communities, students in the CTC
intervention communities will have a reduced likelihood of being an experimenter or alcohol
user compared to being an abstainer, assuming that these classes are replicated.

Methods
34
Data were obtained from students in the Community Youth Development Study (CYDS),
an RCT testing the effectiveness of the CTC model in 24 small to moderatesized
communities (12 matched pairs) located in seven states (Figure 1). One community from each
pair was randomized to implement CTC and the other to conduct prevention services as usual.
From spring 2003 through spring 2008 (Years 15, the efficacy phase of the trial), each of the
 12 intervention communities was provided with training in the CTC model and proactive
technical assistance via weekly phone calls and one to three annual site visits by research staff
to ensure faithful delivery of the CTC model and selected preventive interventions. They also
received funding for a CTC coordinator and up to $75,000 in Years 25 to implement
programs, policies, and practices showing prior evidence of effectiveness in well-conducted
trials, as identified for communities in the CTC Prevention Strategies Guide
(communitiesthatcare.net), and that targeted fifth- to ninth-grade students and their families.

Communities selected interventions that targeted risk factors reported as elevated by local
youth on the same surveys used in the current analyses. Priority risk factors and
interventions differed between communities. Across communities, 17 different school-,
family-, and community-based interventions (examples include: Life Skills Training,
Guiding Good Choices, and Big Brothers/Big Sisters) were implemented during Years 25,
3538
with an average of three programs per community. The sustainability phase of the
CYDS trial began in 2008; from that point, data were collected to evaluate long-term
changes in student outcomes, but training and technical assistance and financial resources
were no longer provided. Most intervention communities continued to implement CTC and
39
evidence-based programs.
The present study began by examining profiles of problem behaviors among tenth graders in
spring 2004. Although the randomized trial began in spring 2003, intervention communities spent
the first 1518 months receiving training, forming coalitions, and planning.
Intervention programming began in fall 2004. No intervention effects were expected in spring
33
2004, and none were found for eighth graders at that time point. By spring 2010,
intervention communities had been implementing CTC with technical and financial support
from the study for 5 years and without study support for 2 additional years. Eleven of the
39
original 12 CTC coalitions were still active and intervention sites were delivering effective
40
prevention programming at higher rates than control communities.

Procedures
Data were obtained from crosssectional population surveys of students administered in
schools during a classroom period by teachers using standardized anonymous administration
procedures. Teachers received specific instructions to assure anonymity of responses and
signed an agreement to abide by these instructions. Screening criteria were used to exclude
respondents who lied or responded inconsistently. For instance, if students responded never to
lifetime use of cigarettes and indicated smoking one or more cigarettes per day during the
past 30 days on a separate question, their answers were considered inconsistent. Students with
over two inconsistent responses or who reported that they were not honest in completing the
survey were excluded. Because responses were anonymous, passive consent was obtained
from parents and informed assent was obtained from students. The University of
Washingtons Human Subjects Review Committee approved the protocol.

Participants
Cross-sectional data were collected from students in grades six, eight, ten, and 12.
Intervention effects were examined for eighth and tenth grades only. In grade six, reported
levels of substance use and delinquency were too low to expect detectable intervention
 th
effects, and lower response rates made the 12 grade crosssectional samples less
representative of the population. Across data sets, the sample consisted of between 49% and
53% girls. Between 3% and 6% of students self-identified as African American, and 77% to
80% as Caucasian. In the 2004 sample, 11% of surveyed tenth graders identified as Hispanic.
In 2010, when new reporting requirements resulted in a change in the wording of
race/ethnicity items such that Hispanic was designated as a separate category, 23% of eighth
graders and 19% of tenth graders identified as Hispanic.

Measures
41,42
Data were collected using the CTC Youth Survey, which is designed to measure risk and
protective factors, substance use, and delinquency in middle and high school samples. Analyses
used items asking the frequency of alcohol, tobacco, and marijuana use across the lifetime and in
the last 30 days, and the number of times in the last 2 weeks they had more than five drinks at one
time. Analyses included three levels of substance use: no use, lifetime but not current use, and
use within the last 30 days. Students also indicated the frequency with which they engaged in
eight delinquent behaviors over the last year (Table 1); these items and binge drinking were
coded yes/no in analyses.

Three community-level variables were included to reduce community-level variability in

outcomes. The percentage of students receiving free or reduced-price lunch and the total
number of students in the community in thousands were obtained from the National Center
for Educational Statistics for the 20032004 school year. Additionally, an indicator for Utah
was included because these students were uniformly lower in levels of substance use. When
community matching was explicitly modeled, the Utah adjustment was unnecessary.

Data Analysis
43
In 2012, multilevel latent class analysis (MLCA) estimated in Mplus was used to examine the
effects of CTC on the probability of students membership in latent classes. The multi-level
component of the model accounted for the fact that communities rather than individual subjects
44,45 33
were randomized. The least restrictive, fully random approach was used to account for
clustering between communities in the probability that a student would be in each latent class
versus the reference class (abstainer was the reference class in this case). Because communities
rather than individuals were randomized to condition, fixed and random effects for the probability
of membership in each class were included to maintain the nominal type I error rate. Full
information maximum likelihood estimation allowed for the inclusion of subjects with missing
data, except those missing data on all variables.

Because the meaning of latent classes may change across time and samples, it is important
to also verify that classes are replicated with independent data; results of these analyses are
only valid in populations where these latent classes exist. Previous analyses of eighth grade
students found support for both four and five-class solutions in independent samples
33
collected in communities from the same states. For this study, the five-class solution was
also replicated for an independent sample of tenth graders (Appendix A).

Initial analyses found that the differences between communities in the distribution of students
across classes were greater than those shown in previous analyses as indicated by intraclass
correlation coefficients (ICCs) for latent class means ranging from of 0.05 to 0.10. This is
 problematic because higher ICCs for the outcome(s) lower statistical power to find treatment
46,47
effects. Community-level covariates were included in the analyses to account for these
differences. The meaning of the latent classes was virtually unchanged, but the ICCs for class
means decreased to between 0 and 0.03, due primarily to communities in Utah compared to other
states. Rates of substance use and problem behaviors were much lower in Utah communities and
thus there were far more abstainers (Table 2).

Results
Initial analyses tested the equivalence of intervention and control conditions for tenth grade
students in 2004, before program implementation. The five-class model including abstainer,
experimenter, alcohol user, drug user, and problem behavior classes found previously in eighth
grade an independent tenth grade sample (Appendix A) was replicated in 2004. No
evidence was found for pre-intervention differences between intervention and control
communities in profiles of substance use and delinquency.

The first step in assessing intervention effects on latent classes was the replication of the
previously identified latent classes. The four-class solution was replicated for eighth graders
in 2010. However, the five-class solution did not distinguish an alcohol use profile, but
instead contained a small group (6%) of students who reported current marijuana use and no
binge drinking (Appendix A). Rather than use this result that was not replicated in any other
sample and therefore potentially unreliable, we proceeded by examining intervention effects
for eighth grade students using four classes, including abstainers, experimenters, drug users,
and problem students, which were substantively the same as the four previously identified
classes. For tenth grade students, the five-class solution including the alcohol class was
replicated in 2010 and was utilized in subsequent analyses. Response profiles for each grade
are shown in Table 1.

Effects of the CTC intervention on class membership in 2010 for both eighth and tenth grades
are reported in Table 2. We hypothesized that CTC would reduce the probability of being in
the experimenter and alcohol use classes versus the abstainer classes for students in the
intervention compared to the control condition. Results showed no intervention effects for the
experimenter class for either grade. For tenth grade students, the intervention reduced the
probability of being in the alcohol use versus the abstainer class, as hypothesized. The odds of
a tenth grade student in an intervention community being in the alcohol use class as opposed
to the abstainer class was 0.69 (CI=0.48, 1.00) times those of a student in the control
condition. To show the practical implication of these effects, the aforementioned multinomial
regression model was used to calculate the predicted probability that a student from a control
community not in Utah, at the average level of free and reduced lunch (39%), and of the
average student population size (3,417 students) would be an alcohol user was 0.26 versus
0.19 if the student lived in a treatment community, the expected reduction in alcohol users
due to CTC is from 26% to 19%. Analyses showed no difference in the probability of
belonging to either the drug use or problem behavior classes versus the abstainer class in the
CTC intervention communities compared to the control communities in eighth or tenth grade
crosssectional samples (Table 2).

Discussion
CTC was designed to reduce levels of substance use, delinquency, and other problem
behaviors using a communitybased prevention system that relies on broad-based coalitions
to reduce elevated risk factors and strengthen protective factors with effective preventive
interventions. Previous studies have shown significant intervention effects in smoking,
drinking, and delinquency in a longitudinal panel of students followed from grade five
22, 23
through ten. Instead of the longitudinal panel, the current study used two cross
sectional samples of eighth and tenth grade students collected 6 years apart: in 2004, before
prevention programs had been initiated, and in 2010, 2 years after external support for CTC
and prevention program implementation had been withdrawn but while intervention sites
were continuing to deliver effective prevention programming. The current study differed
from prior outcome analyses in that it tested the specific hypothesis that CTC would reduce
the proportion of students in the experimenter and alcohol user classes in repeated cross
sectional surveys. Because some universal interventions have been shown to have the
2630
strongest effects among the highest-risk youth, the study also assessed the degree to
which intervention effects were found among youth engaging in more serious or higher
levels of problem behaviors. This possibility was tested by default, given that differences in
all profiles were assessed with the LCA approach.

Only one of three hypothesized intervention effects was found: a 7percentage point
reduction in alcohol users in tenth grade. Contrary to our hypothesis, no differences in
experimenters were found across intervention conditions for either eighth or tenth grade
students in the crosssectional 2010 survey. Analyses showed no evidence of intervention
effects on more serious levels of problem behaviors, although these analyses were
exploratory and effects were not specifically hypothesized. The inclusion of community-
level covariates substantially reduced variability in latent classes between communities
because the students in Utah were much more likely to be abstainers than students in any
other state. These conclusions assume the stability of the four- and five-class models in the
population.

The analyses utilized in this study allowed identification of a group of students for whom the
intervention was most effective: tenth graders who displayed a specific profile of being likely
to engage in current drinking and binge drinking, but not other substance use or problem
behaviors. Although it is unclear why intervention effects would be largest on alcohol users,
22,23
that effects on alcohol were found in both cross-sectional and longitudinal data is
noteworthy. Cross-sectional analyses typically have less power to detect intervention effects
48
than analyses of longitudinal data. The reduction in rates of alcohol use is also important
49
because alcohol is the most commonly used substance among adolescents. Moreover, use of
alcohol during adolescence has been associated with an increased risk of other negative
outcomes not measured in this survey, including victimization, risky sexual behavior, and
25,5052
mental health problems, as well as an increased likelihood of alcohol abuse and
25,53
dependency during adulthood.
An unexpected finding was that the intervention did not affect the proportion of students
who were experimenters in either grade and that the alcohol use class was not replicated in
the eighth grade. One explanation for these null findings is the systematic decrease in
substance use seen both in these data (the percentage of eighth grade abstainers increased
 49
from 58% in 2004 to 66% in 2010) and in recent years in national surveys. These trends
could have impacted profiles of substance use such that the alcohol use class was no longer
evident for eighth graders. Decreasing percentages of experimenters in eighth and tenth
grade also reduces power to find intervention effects for these students. Future cross
sectional surveys of students in the CTC study communities could help inform some of
these issues. Although this study did not look at intervention effects in other years, it would
be especially informative to examine these effects in 2008 and, given that interventions are
sustained, in 2012 data.
This study did not find a reduction in rates of serious drug use or problem behaviors for
students in intervention versus control communities. These analyses also had low power to
detect effects given the low rates of involvement in these behaviors. Additionally, given that
communities largely implemented universal prevention strategies targeting youth who had
not yet initiated involvement in problem behaviors, we did not have strong expectations for
finding intervention effects in these groups.

Although this paper did not demonstrate the hypothesized intervention effects, the more
nuanced evaluation provided by these analyses did allow identification of a specific
subgroup of students for whom the intervention was effective. The strength of this LCA
approach is that it recognizes that not all substance or alcohol use is the same, and allows
for the creation of particular profiles of problem behaviors based on all measured outcomes.
In this case, the analyses were able to distinguish between alcohol users, those who
experimented with substance use, and those who used alcohol as part of a pattern of more
serious problem behaviors. It is possible that students who engage only in alcohol use view
this behavior as normative, whereas they consider other illegal drugs or delinquent
behaviors as more deviant. The current study suggests that CTC may be effective at altering
the attitudes and behaviors of alcohol users more than those of others.