PARENTAL MONITORING AND CANNABIS USE: EPIDEMIOLOGICAL
EVIDENCE FROM TWO PROSPECTIVE COHORT STUDIES
By
Kipling Morgan Bohnert

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Epidemiology
2010

ABSTRACT
PARENTAL MONITORING AND DRUG USE: EPIDEMIOLOGICAL EVIDENCE
FROM TWO PROSPECTIVE COHORT STUDIES
By
Kipling Morgan Bohnert
Background: Cannabis is one of the most commonly used psychoactive drugs in the
world. The adept level of parental monitoring during childhood and adolescence is one
potential intervention target for preventing cannabis use or delaying its onset.
Aims: The overall aim of this dissertation is to estimate the potential short- and long-term
impact of parental monitoring on cannabis. The first study (Chapter Three) aims to
estimate, prospectively, the influence of parental monitoring assessed at age 11 on
cannabis initiation before age 18 years. Next, (Chapter Four) the goal is to estimate the
suspected influence of level of parental monitoring assessed at age 11 on level of drug
use at age 17 years, while simultaneously testing paths through levels of drug use and
affiliation with drug using peers at age 11 years. The final study (Chapter Five) aims to
test the prospective association that might link level of parental monitoring with
subsequent recently active cannabis smoking, and to examine a potential meditational
influence of level of deviant peer affiliation.
Methods: Data for the first two studies are from a longitudinal study of a 1983-1985 birth
cohort from southeast Michigan (n=823). Data for the third study come from a
prospective cohort of students from an urban public school system in the mid-Atlantic
United States entering primary school in 1985 and 1986 (n=2,311). For all three studies
drug use was assessed via standardized, self-reported measures and parental monitoring

was assessed via a 10-item standardized child-reported scale. In the first study, Poisson
regression with robust error variances was used to estimate the suspected predictive
association that links parent monitoring at age 11 and cannabis use up to age 17. The
next study used a structural equation modeling (SEM) approach to estimate paths of
parental monitoring influence, with levels of affiliation with drug using peers and drug
use modeled as latent variables. The final study applied a generalized estimating
equations (GEE) logistic regression approach to estimate the association between prior
parental monitoring and subsequent recent cannabis use. SEM was used to examine the
potential mediating influence of level of deviant peer affiliation.
Results: In the first study, higher parental monitoring at age 11 signaled a reduced risk of
cannabis initiation from ages 11 to 17 years (adjusted estimated relative risk = 0.96; 95%
confidence interval = 0.94, 0.98). Results from the second study indicated that level of
parental monitoring was related to levels of affiliation with drug using peers (p<0.05) and
drug use at age 11 (p<0.05) and predicted levels of drug use at age 17 (p<0.05). In the
third study, higher levels of prior parental monitoring significantly predicted lower odds
of recently active cannabis use (adjusted odds ratio (AOR) = 0.96; 95% confidence
interval (CI) = 0.92, 0.99). Level of deviant peer affiliation did not appreciably mediate
the association between previous level of parental monitoring and subsequent cannabis
use.
Conclusions: These findings from prospective research shed new light and help confirm
the theory that parenting and familial characteristics might exert long-lasting influences
on a child’s risk of initiation and use of illegal drugs. Implications for prevention and
future directions for research are discussed.

Copyright by
KIPLING MORGAN BOHNERT
2010

ACKNOWLEDGEMENTS

Although a dissertation is an individual project, it cannot be completed without
the direct and indirect support of others. Therefore, I am deeply indebted to a number of
people.
I would like to thank my wife and partner, Amy Bohnert, for her unconditional
love, support, and patience, especially over the course of completing this dissertation. I
am inspired by you and am constantly learning from you.
I would also like to thank other members of my family. I am grateful for the love
and generosity of my parents, Thom and Michele Bohnert. Thank you for providing me
with a warm and enriching environment, especially during my formative years, instilling
in me a deep sense of curiosity and an appetite for learning, and making education a
priority. I would also like to thank my sister, Jemere Bohnert, for her dependable
friendship throughout the years. I would also like to thank my grandparents for being
positive role models and showing me the value of a positive work ethic. I would also like
to thank the Buchanans for their generosity, love, and support.
I would like to express my sincere gratitude to my advisors and dissertation
committee members. I wish to thank Jim Anthony for his guidance and mentorship
during the course of my dissertation. I have gained tremendous epidemiological and
methodological knowledge from working with you, which I will carry with me. I am also
grateful for your, and S.B.’s, warm hospitality during my graduate education. I wish to
thank Naomi Breslau for her support during my studies. I have taken note of your
v

intellectual rigor, including the clarity of your thinking and insightful criticism, which I
will take with me throughout my career. I would also like to thank Jim and Naomi for
providing me with the data upon which this dissertation is based. I wish to thank Qing
Lu for his methodological advice and for providing insightful feedback on my
dissertation. I would also like to thank Siddharth Chandra for taking time to meet with
me, serve on my dissertation committee, and giving thoughtful comments on my
dissertation. I also wish to thank the other faculty members in the Department of
Epidemiology for the knowledge that I have acquired from them, as well as their support,
during my time as a student.
This research was made possible by the following grants from NIDA and NIMH:
F31DA021040, MH44586, MH71395, DA09897, DA04392, DA019805, and
K05DA015799.

vi

TABLE OF CONTENTS

LIST OF TABLES…………………………………………………………………….......x
LIST OF FIGURES……………………………………………………………………...xii
CHAPTER ONE
STATEMENT OF THE PROBLEM AND SPECIFIC AIMS…………………………...1
1.1 Statement of the Problem..…………………………………………………….1
1.2 Specific Research Aims……..………………………………………………...5
CHAPTER TWO
REVIEW OF THE LITERATURE……………………………………………………….7
2.1 Epidemiology of Cannabis Use …………………….………………………...7
2.1.1 Structure of the Review of the Literature on the Epidemiology of
Cannabis Use……………………………………………………………...7
2.1.2 Quantity ……………………………………..……………………...8
2.1.2.1 Worldwide Cannabis Use……………………………..…..8
2.1.2.2 Cannabis Use in the United States………………………..9
2.1.3 Location……………………………………………………………10
2.1.3.1 Individual and Sociodemographic Characteristics……….10
2.1.3.2 Variation by Place………………………………………..11
2.1.3.3 Time Trends……………………………………………...12
2.1.4 Suspected Causes…………………………………………………..13
2.1.4.1 Family History and Genetics…………………………….13
2.1.4.2 Parenting Practices……………………………………….15
2.1.4.3 Affiliation with Deviant and Drug Using Peers………….16
2.1.4.4 Intra-Individual Traits and Preexisting Psychiatric
Problems………..………………………………………………..17
2.1.5 Mechanisms………………………………………………………..18
2.1.5.1 The Gateway Hypothesis and Exposure Opportunity……18
2.1.5.2 Cannabis Dependence……………………………………20
2.1.5.3 Educational Outcomes…………………………………...21
2.1.5.4 Other Psychiatric Consequences…………………………22
2.1.6 Prevention and Control…………………………………………….23
2.2 Prior Research on Parental Monitoring………………………………………24
2.2.1 Definition of Parental Monitoring…………………………………25
2.2.2 Parental Monitoring Conceptualized within a Developmental Model
for Antisocial Behavior………………………………………………….25
2.2.3 General Conceptual Models Linking Parental Monitoring with
Cannabis Use…………………………………………………………….30
2.3 Conclusions…………………………………………………………………..33

vii

CHAPTER THREE – MANUSCRIPT ONE
PARENTAL MONITORING AT AGE 11 AND SUBSEQUENT ONSET OF
CANNABIS USE UP TO AGE 17: RESULTS FROM A PROSPECTIVE STUDY…..34
Abstract…………………………………………………………………………..34
3.1 Introduction …………………………………….............................................36
3.2 Methods………………………………………………………………………38
3.2.1 Sample …………………………………………………………..…38
3.2.2 Measures …………………………………………………………..39
3.2.2.1 Cannabis Initiation, Ages 11-17 Years…………………..39
3.2.2.2 Parental Monitoring, Age 11 Years ……………………..40
3.2.2.3 Child Covariates………………………………………….40
3.2.2.4 Maternal Covariates……………………………………...41
3.2.3 Statistical Analysis…………………………………………………41
3.3 Results………………………………………………………………………..42
3.3.1. Comparison of the Original Sample with the Follow-up Subsamples at Ages 11 and 17 Years…………..……………………………42
3.3.2 Parental Monitoring at Age 11 and the Risk of Cannabis Initiation up
to Age 17…………………………………………………………………45
3.3.3 Kaplan-Meier Survival Estimates………………………………….49
3.3.4 Fractional Polynomial Post-estimation Examination of the Parental
Monitoring – Cannabis Initiation Association…………………………...51
3.4 Discussion……………………………………………………………………53
CHAPTER FOUR – MANUSCRIPT TWO
LONGITUDINAL STUDY OF LEVELS OF ADOLESCENT DRUG USE: AN SEM
ANALYSIS OF PARENTAL MONITORING………………………………………….58
Abstract…………………………………………………………………………..58
4.1 Introduction ………………………………………………………………….59
4.2 Methods………………………………………………………………………61
4.2.1 Data and Sample …………………………………………………..61
4.2.2 Measures …………………………………………………………..62
4.2.2.1 Parental Monitoring ……………………………………..62
4.2.2.2 Drug Use ………………………………………………...63
4.2.2.3 Affiliation with Drug Using Peers……………………….64
4.2.2.4 Baseline Covariates………………………………………64
4.2.3 Statistical Analysis…………………………………………………64
4.3 Results………………………………………………………………………..65
4.3.1. Comparison of the Original Sample with the Subset Used in the
Analysis…………………………………………………………………..65
4.3.2 Mean Levels of Parental Monitoring and Percentages of Drug Use in
the Sample………………………………………………………………..67
4.3.3 Structural Equation Model for Parental Monitoring, Drug Use, and
Peer Drug Use at Age 11 and Drug Use at Age 17………………………69
4.4 Discussion……………………………………………………………………73

viii

CHAPTER FIVE – MANUSCRIPT THREE
LONGITUDINAL PATHS FROM PARENTAL MONITORING TO CANNABIS USE
IN CHILDHOOD AND ADOLESCENCE……………………………………………...77
Abstract…………………………………………………………………………..77
5.1 Introduction …………………………………….............................................79
5.2 Methods………………………………………………………………………81
5.2.1 Study Population and Sample……………………………………...81
5.2.2 Main Measures ………………………………………………….....83
5.2.2.1 Parental Monitoring, PM ………………………………..83
5.2.2.2 Recently Active Cannabis Smoking, CAN ……………...83
5.2.2.3 Deviant Peer Affiliation, DPA…………………………...84
5.2.3 Other Covariates…………………………………………………...84
5.2.4 Statistical Analysis…………………………………………………85
5.3 Results………………………………………………………………………..88
5.3.1 Characteristics of the Sample………………………………………88
5.3.2 Prevalence of CAN and Mean Scores for PM and DPA, 19891994………………………………………………………………………90
5.3.3 Results from the GEE Logistic Regression on the Hypothesized PMCAN Link………………………………………………………………...92
5.3.4 Examining Clustering at the Individual and Classroom/school Levels
Via the ALR……………………………………………………………...95
5.3.5 Testing Hypothesized Mediation through DPA……………………97
5.4 Discussion…………………………………………………………………..105
CHAPTER SIX
FINAL CONCLUSIONS………………………………………………………………109
6.1 Summary of Findings……………………………………………………….109
6.2 Limitations………………………………………………………………….110
6.3 Future Directions…………………………………………………………...112
6.4 Conclusions…………………………………………………………………114
APPENDICIES
Appendix A:
Appendix B:
Appendix C:
Appendix D:
Appendix E:

Technical Appendix for Chapter Three………………………...116
Technical Appendix for Chapter Four………………………….120
Technical Appendix for Chapter Five…………………………..123
Parental Monitoring Scale………………………………………129
Deviant Peer Affiliation Scale …………………………………130

REFERENCES
References……………………………………………………………………....132

ix

LIST OF TABLES

Table 3.1

Description of the initial sample, the subset with follow-up data for ages
11 and 17, and the subset used in the regression analyses. Data come from
823 children sampled from 1983-1985 newborn discharge lists in
southeast Michigan and assessed at ages 6, 11, and 17 years……………44

Table 3.2

Parent monitoring at age 11 and initiation of cannabis use up to age 17,
unadjusted models. Data come from 641 children sampled from 19831985 newborn discharge lists in southeast Michigan with complete
information on all variables……………………………………………...47

Table 3.3

Parent monitoring at age 11 and initiation of cannabis use up to age 17,
adjusted model. Data come from 641 children sampled from 1983-1985
newborn discharge lists in southeast Michigan with complete information
on all variables…………………………………………………………...48

Table 4.1

Description of the initial sample and the subset used in the structural
equation model (SEM). Data come from 823 children sampled from
1983-1985 newborn discharge lists in southeast Michigan and assessed at
ages 6, 11, and 17 years………………………………………………….66

Table 4.2

Mean level of parental monitoring and percentages of drug use in the
sample. Data come from 774 children sampled from 1983-1985 newborn
discharge lists in southeast Michigan and contributing data to the
analyses………………………………………………………………….68

Table 5.1

Characteristics of the sample at each assessment 1989-1994. Data come
from 2,311 first graders enrolled in a mid-Atlantic public school system
and assessed yearly from 1989 through 1994……………………………89

Table 5.2

Percentages of recent cannabis use and means scores of parental
monitoring and deviant peer affiliation for the assessments. Data come
from a cohort of 2,311 first graders enrolled in a mid-Atlantic public
school system and assessed yearly from 1989 through 1994………..…..91

Table 5.3

Results from the unadjusted GEE linking prior parental monitoring with
odds of recent cannabis use. Data come from 1,448 first graders enrolled
in a mid-Atlantic public school system and assessed yearly from through
1994……………………………………………………………….……..93

Table 5.4

Results from the adjusted GEE logistic regression linking prior parental
monitoring with odds of recent cannabis use. Data come from 1,448 first
graders enrolled in a mid-Atlantic public school system and assessed
yearly from 1989 through 1994………………………………………….94
x

Table 5.5

Results from the ALR linking prior parental monitoring with odds of
recent cannabis use and accounting for clustering. Data from 1,448
participants originally recruited in 1985-1986 and followed-up from 19891994………………………………………………………………………96

Table 5.6

Formal tests of mediation for the two main paths of interest that were
robust in the SEM. Data come from a cohort of 2,311 first graders enrolled
in a mid-Atlantic public school system and assessed yearly from 1989
through 1994……………………………………………………………104

xi

LIST OF FIGURES

Figure 1.1

The general conceptual model linking parental monitoring with drug
use…………………………………………………………………………4

Figure 2.1

The developmental model for antisocial behavior……………………….27

Figure 2.2

Re-conceptualization of the developmental model for antisocial
behavior…………………………………………………………………..29

Figure 2.3

The general conceptual model linking parental monitoring with drug
use………………………………………………………………………..31

Figure 2.4

The general conceptual model linking parental monitoring with drug use
extended to include possible mediation, subgroup variation, and other
possible explanatory variables…………………………………………...32

Figure 3.1

Kaplan Meier estimates of cumulative incidence of cannabis use from age
11 to 17 years. Data come from 619 children sampled from 1983-1985
newborn discharge lists from two hospitals in southeast Michigan……..50

Figure 3.2

Estimated probability of cannabis initiation from ages 11 to 17 years by
level of parental monitoring. Data come from 641 children sampled from
1983-1985 newborn discharge lists in southeast Michigan with complete
information on all variables……………………………………………...52

Figure 4.1

Unadjusted SEM predicting levels of drug use at age 17 (Drug 17) from
levels of drug use (Drug 11), parental monitoring (PM 11) and affiliation
with drug using peers at age 11 (Drug11). Data come from 774 children
sampled from 1983-1985 newborn discharge lists in southeast Michigan
and contributing data to the analyses……………………………………70

Figure 4.2

Adjusted SEM predicting levels of drug use at age 17 (Drug 17) from
levels of drug use (Drug 11), parental monitoring (PM 11) and affiliation
with drug using peers at age 11 (Drug11). Data come from 774 children
sampled from 1983-1985 newborn discharge lists in southeast Michigan
and contributing to the analyses…………………………………………72

Figure 5.1

SEM estimates linking prior level of parental monitoring (PM) with recent
cannabis use (CAN). Data come from 1,302 first graders enrolled in a
mid-Atlantic public school system and assessed yearly from 1989 through
1994………………………………………………………………………98

xii

Figure 5.2

SEM estimates linking prior level of parental monitoring (PM) with level
of deviant peer affiliation (DPA). Data come from 1,302 first graders
enrolled in a mid-Atlantic public school system and assessed yearly from
1989 through 1994……………………………………………………...100

Figure 5.3

SEM estimates testing paths from level of parental monitoring (PM) to
level of deviant peer affiliation (DPA) to recent cannabis use (CAN).
Data come from 1,302 first graders enrolled in a mid-Atlantic public
school system and assessed yearly from 1989 through 1994…….…….102

xiii

1. CHAPTER ONE

STATEMENT OF THE PROBLEM AND SPECIFIC AIMS

1.1 Statement of the Problem

Cannabis is one of the most widely used psychoactive drugs in the world
(Degenhardt et al., 2008). Cumulative occurrence of cannabis use is especially high in
the United States. In 2008, for example, over 40% of individuals aged 12 and older had
tried cannabis at least once in their lives, with an estimated 2.2 million of them using for
the first time within 12 months of the date of assessment (Substance Abuse and Mental
Health Service Administration, 2009). Moreover, estimates from the National
Comorbidity Survey have projected that nearly one out of every ten individuals who
smoke cannabis eventually develop the clinical syndrome of cannabis dependence (James
C. Anthony, Warner, & Kessler, 1994). There is also a body of evidence on psychiatric
complications of cannabis smoking, the most disabling of which involves schizophrenia
or schizophrenia-like illnesses (Morrison et al., 2009; Tien & Anthony, 1990).
The use of cannabis in childhood and adolescence is of particular public health
salience. Compared with cannabis users who initiate during adulthood, early cannabis
users have been found to experience excess problems related to their use, namely DSMIV drug abuse or dependence (Chen, O'Brien, & Anthony, 2005). In addition, early
cannabis users have been observed to be more likely to go on to use other drugs (D. B.
Kandel, 1984; Yamaguchi & Kandel, 1984a), possibly due to increased opportunities to

1

try other drugs (Wagner & Anthony, 2002b). Recent epidemiological evidence has also
shown a link between early cannabis use and later adverse health and social outcomes.
These adverse outcomes include educational outcomes and other psychiatric disturbances
(Andreasson, Allebeck, Engstrom, & Rydberg, 1987; Fergusson & Boden, 2008a; M.
Lynskey & Hall, 2000).
Therefore, it is important to identify potential factors that might prevent or delay
cannabis use. One of the potential factors that has received attention is parental
monitoring. Parental monitoring is one facet of parenting, and is defined chiefly by
parental supervision, tracking and knowledge of child whereabouts (Dishion &
McMahon, 1998). The guiding conceptual framework for the link between parental
monitoring and drug use and other antisocial behavior has been presented by Patterson
and colleagues (Patterson, DeBaryshe, & Ramsey, 1989) and later refined by Dishion and
McMahon (Dishion & McMahon, 1998). The developmental progression for antisocial
behavior, sometimes referred to as the “social context model”, posits that parental
monitoring reduces drug use in part by reducing affiliation with deviant peers.
Despite an accumulation of empirical evidence and a well-articulated conceptual
framework to link parental monitoring with later drug use initiation in children and
adolescents (Dishion & McMahon, 1998), gaps in the literature remain. First, few
epidemiologic studies have examined cannabis initiation exclusively in the parental
monitoring-drug initiation relationship. Most prior research has grouped all drugs
together or has only examined tobacco or alcohol initiation. However, cannabis use in
childhood and adolescence is important for the reasons described in the preceding
paragraphs. Second, previous studies on parental monitoring and drug use have not been

2

exhaustive in coverage of potentially important explanatory covariates. Third, it is
uncertain whether the association between parental monitoring and drug use is uniform
across all subgroups of the population (e.g., across subgroups defined by sex and
race/ethnicity). Fourth, prior investigations have been limited in their examination of the
link between parental monitoring and subsequent health outcomes to a relatively short
time-span, usually no more than two or three years. Fifth, to date, no study on cannabis
has tested the main components of the social context model simultaneously within a
longitudinal context; that is, tested the paths from earlier parental monitoring to later
affiliation with deviant peers to later cannabis use.
This dissertation research will seek to fill in these existing gaps in evidence by
examining the role of parental monitoring in cannabis initiation and use. The research is
based on longitudinal data from two cohort studies. One study sample consists of a
cohort of 823 children born in 1983-1985 in southeast (SE) Michigan. Baseline data
were gathered from mothers and children when the children were 6 years of age. Two
additional follow-up assessments took place when the cohort members were 11 and 17
years of age. A second study sample is from a cohort study of 2311 youths who were
enrolled in first grade in 19 primary schools in an urban public school system in the midAtlantic United States during two successive school years in 1985–1986. Follow-up
assessments were conducted every year until 1994.
Data from these sources will allow for the estimation of the hypothesized
relationships that link parental monitoring with later occurrence of cannabis smoking and
other drug use, within the framework of the general conceptual model depicted in Figure
1.1.

3

Figure 1.1 The general conceptual model linking parental monitoring with drug use

Parental
monitoring

Drug use

4

1.2 Specific Research Aims

Three studies in the form of manuscripts will be conducted for the dissertation
research, with the following specific aims, each of which involves estimation of specific
relationships or paths within the overarching framework of the general conceptual model.

Specific Aim 1: Using data from the SE Michigan cohort, to estimate
prospectively the suspected causal association that links parental monitoring in
pre-adolescence (age 11) with the cumulative incidence of cannabis use initiation
up to late adolescence (age 17), the age interval when cannabis initiation is most
likely to occur.
Sub Aim 1a: To test whether the relationship varies between blacks and
whites, and/or males and females, or whether a single regression slope
estimate serves well to summarize this association.
Sub Aim 1b: To study the time to first cannabis use in relation to level of
parental monitoring.

Specific Aim 2: Using data from the SE Michigan cohort, to estimate the
influence of level of parental monitoring assessed at age 11 on levels of drug use
at age 17, while simultaneously examining potential paths through levels of drug
use and deviant peer affiliation at age 11.

5

Specific Aim 3: Using data from the Mid-Atlantic cohort study, to estimate the
suspected influence of earlier levels of parental monitoring on later cannabis use
over a five year interval.
Sub Aim 3a: To test whether the association is consistent over the five
year interval or whether it varies as a function of age.

Specific Aim 4: Using data from the Mid-Atlantic cohort study, to simultaneously
test the suspected paths linking earlier levels of monitoring to subsequent levels of
affiliation with deviant peers to later cannabis use.

All of these aims involve specification of a statistical model that is appropriate for
estimation.

6

2. CHAPTER TWO

REVIEW OF THE LITERATURE

2.1 Epidemiology of Cannabis Use

2.1.1 Structure of the Review of the Literature on the Epidemiology of Cannabis Use

The present review of the literature on the epidemiology of cannabis use is
organized in relation to the five main rubrics of epidemiology, as suggested by (J. C.
Anthony & Van Etten, 1998). Accordingly, the sequence of sub-headings of this section
is as follows: (1) Quantity, with respect to the global burden of disease associated with
cannabis use; (2) Location, with respect to subgroups of the population that are
differentially affected by cannabis use, and the location of affected subgroups as defined
in relation to characteristics of person, place, and time; (3) Suspected Causes, with
respect to characteristics that have come to be accepted as possible causal determinants of
cannabis use; (4) Mechanisms, with respect to linkages of states and processes that lead
up to cannabis use and that influence its consequences or aftermath (e.g., secondary comorbidities and disabilities or impairments if the condition is not treated effectively); and
(5) Prevention and Control, with respect to mass action interventions that might be used
to reduce incidence, duration, or suffering associated with cannabis use. The review
includes the significance of the cannabis use in terms of health and social outcomes and
has a particular emphasis on early cannabis use (i.e., use in childhood and adolescence).

7

2.1.2 Quantity

2.1.2.1 Worldwide Cannabis Use

Globally, cannabis is one of the most commonly used psychoactive drugs.
According to estimates from the United Nations, between 129 and 191 million people
used cannabis in 2008; equivalent to 2.9% to 4.3% of the entire global population aged
15 to 64 years (UNODC, 2010). In 2008, Oceania, which includes Australia and New
Zealand, and the Americas had the highest annual prevalence; 9.3% to 14.8% and 6.3%
to 6.6% of their populations aged 15 to 64, respectively (UNODC, 2010). Furthermore,
within these regions with a high prevalence of cannabis use, there was substantial
variation within region. For example, within the Americas, North America had a higher
percentage of past year cannabis use (~10%) compared with South America (~3%)
(UNODC, 2010).
Estimates of the cumulative incidence or cumulative occurrence (sometimes
called “lifetime prevalence”) of cannabis use have been published recently by the World
Mental Health (WMH) consortium (Degenhardt et al., 2008). Cumulative incidence
varied widely among the study sites, ranging from less than 1% in the People’s Republic
of China to 42.4% in the United States (US) (Degenhardt et al., 2008). European nations
had relatively moderate estimates of cumulative incidence (ranging from 6.4% to 19.0%);
the Middle East and Africa had relatively low percentages (ranging from 2.7% to 8.4%)
(Degenhardt et al., 2008).

8

2.1.2.2 Cannabis Use in the United States

Turning to evidence from the US, findings from the 2008 National Survey on
Drug Use and Health (NSDUH) indicate that an estimated 41% of individuals aged 12
and older have used cannabis at least once in their lives (Substance Abuse and Mental
Health Service Administration, 2009). This estimate is consistent with the one obtained
by the WMH. According to the NSDUH, an estimated 15.2% of the US household
population were current (past-month) cannabis users (Substance Abuse and Mental
Health Service Administration, 2009).
The NSDUH also provides estimates for annual incidence, which might have less
of the biases associated with estimating cumulative occurrence of use (e.g., older adults
having to recall cannabis use earlier in life). In the 2008 NSDUH, an estimated 2.2
million individuals started to use cannabis (Substance Abuse and Mental Health Service
Administration, 2009). This number is equivalent to about 0.9% of the total US
household population, and 1.5% of those at risk for initiation (i.e., among individuals
who have never tried cannabis). The number of recent initiates obtained in 2008, i.e.,
around 2 million individuals, was similar to estimates obtained for the five preceding
years of the NSDUH (Substance Abuse and Mental Health Service Administration,
2009). Further, more than 60% of the new initiates in 2008 were under the age of 18
years when they first tried cannabis (Substance Abuse and Mental Health Service
Administration, 2009), indicating that cannabis initiation is most likely to occur at
younger ages.

9

US population-based school surveys, such as Monitoring the Future (MTF),
provide additional evidence with respect to the rubric of quantity. In the 2008 MTF
survey, an estimated 15% of eighth-graders, 30% of tenth-graders, and 43% of twelfthgraders had ever smoked cannabis (Johnston, O'Malley, Bachman, & Schulenberg, 2009).
Taken together, these estimates illustrate that cannabis use in the US is relatively
common and usually begins before adulthood.

2.1.3 Location

2.1.3.1 Individual and Sociodemographic Characteristics

There are several sociodemographic characteristics associated with cannabis use,
and cannabis initiation, in particular. As previously described in the section on quantity,
age is one of the most noteworthy characteristics associated with cannabis use and
initiation. Moreover, as reported by the 2008 NSDUH, the mean age at first cannabis use
was 17.8 years among recent initiates aged 12 to 49 years (Substance Abuse and Mental
Health Service Administration, 2009). In addition, it has been noted that almost twothirds of new initiates are under the age of 18 years (Substance Abuse and Mental Health
Service Administration, 2009). Few individuals initiate cannabis use after the age of 26
years (Substance Abuse and Mental Health Service Administration, 2009).
With respect to sex, it has been commonly observed that males have an excess
risk of cannabis initiation (Substance Abuse and Mental Health Service Administration,
2009). These differences might be traced back to exposure opportunity; that is, males

10

have been found to be more likely to be offered a chance to try cannabis than females, but
given this chance to try both sexes are equally likely to go on to use (Van Etten &
Anthony, 1999). There is evidence, however, that male-female differences with respect
to initiation might be less substantial in recent years (Substance Abuse and Mental Health
Service Administration, 2009). Recent evidence from the NSDUH shows virtually no
male-female differences with respect to initiation (Substance Abuse and Mental Health
Service Administration, 2009).
Differences with respect to race/ethnicity have been observed in cannabis
initiation. However, it is often difficult to disentangle these observed associations from
social disadvantage. For example, rates of annual cannabis incidence are much higher for
American Indians/Alaskan Natives (J. C. Gfroerer, Wu, & Penne, 2002; Substance Abuse
and Mental Health Service Administration, 2009); however, American Indians/Alaskan
Natives also tend to have lower levels of socioeconomic status.

2.1.3.2 Variation by Place

As presented in the rubric of quantity, there is substantial variation in cannabis
use by region and country of the world. It should be noted that it is often difficult to
compare cannabis consumption patterns, mainly due to the variation in resources
available to collect data. As noted in the UN report, many countries around the world
lack adequate resources to conduct large-scale epidemiologic studies of drug use
(UNODC, 2010). Therefore, reliable estimates of cannabis use are scarce in these
countries. Nonetheless, it is generally observed that countries with relatively higher

11

levels of per capita income have higher percentages of recent (past-year) and cumulative
(lifetime) use (Degenhardt et al., 2008; UNODC, 2010). Highest percentages of cannabis
use have been consistently noted for the US, New Zealand, and Australia (UNODC,
2010).
Studies have also shown variation in recent cannabis use at the microsocial level.
For example, studies in the US and New Zealand have found that recent cannabis use
tends to cluster within neighborhoods (Bobashev & Anthony, 1998; Wells, Degenhardt,
Bohnert, Anthony, & Scott, 2009). The estimates are of similar magnitude for both
countries (pair-wise odds ratios between 1.3 and 1.6), and are also similar to the amount
of clustering seen in villages of the developing world during diarrheal disease outbreaks
(Bobashev & Anthony, 1998; Wells et al., 2009).

2.1.3.3 Time Trends

As described previously in the rubric of quantity, the number of annual cannabis
initiates among US households has remained relatively stable over the past five years for
which there is data, an estimated 2 million individuals aged 12 years and older
(Substance Abuse and Mental Health Service Administration, 2009). A report by
Gfroerer and colleagues (2002) provides more information on time trends with respect to
annual incidence in the US since 1965 (i.e., reliable statistics before 1965 are relatively
scarce within the US) (J. C. Gfroerer et al., 2002). The authors found that the rate of
annual cannabis initiation rose from less than 5 incident users per 1000 person-years for
individuals aged 12 years and older during the late 1960s and early 1970s to a peak in

12

1976-1977 of more than 20 incident users per 1000 person-years (J. C. Gfroerer et al.,
2002). Since that peak, cannabis initiation declined to 8.5 per 1000 person-years in 1990,
and then increased steadily again to almost 17 per 1000 person-years in 1996 (J. C.
Gfroerer et al., 2002). After 1996, rates declined to 13.6 per 1000 in 1999 (J. C. Gfroerer
et al., 2002). In the MTF, the prevalence of recent (past-year) use among twelfth graders
peaked in 1978 and 1979 and decreased during the 1980s until 1992, when it doubled
from 1992 to 1997 from an estimated 22% to 39% (Johnston et al., 2009). Since 1997,
annual prevalence has declined and leveled to an estimated 32% (Johnston et al., 2009).
The general trends described above were observed for both males and females;
however, rates for males were higher. The mean age at first use decreased from about 19
years in the early 1970s to 17 years in the 1990s (J. C. Gfroerer et al., 2002). Recent data
from the NSDUH found that the average age of new initiates was just under 18 years
(Substance Abuse and Mental Health Service Administration, 2009). Similarly, evidence
from the cross-sectional WMH surveys found that use was especially common in younger
cohorts (Degenhardt et al., 2008).

2.1.4 Suspected Causes

2.1.4.1 Family History and Genetics

Familial studies have presented evidence for the shared influences of cannabis
use. For example, parent-child correlations of cannabis use have been estimated to range
from 0.3 to 0.5 (Brook, Whiteman, Gordon, & Brook, 1985; J. Gfroerer, 1987).

13

Likewise, correlations between older brothers and younger siblings have been observed
to be of similar magnitude (Brook, Whiteman, Gordon, & Brenden, 1983). In other
epidemiologic research along these lines, the odds of lifetime cannabis use disorders were
substantially elevated in siblings, adult children, and spouses of probands with cannabis
use disorders (odds ratios ranging from 3.6 to 6.9) (Merikangas et al., 2009). These
findings are consistent with previous findings reported by the authors, and of similar
magnitude (Merikangas et al., 1998).
Twin studies have been able to provide further clues about genetic, shared
environmental, and non-shared environmental sources of variance. Kendler and
colleagues have estimated the heritability, i.e., the genetic influence, of cannabis use
using male and female population-based twin registries; the estimates ranged from 0.4 for
females to 0.3 for males (Kendler, Karkowski, Neale, & Prescott, 2000; Kendler &
Prescott, 1998). In these studies, shared and non-shared environmental influences were
roughly equivalent for males (~0.3) but differed for females (0.35 for shared and 0.25 for
non-shared) (Kendler et al., 2000; Kendler & Prescott, 1998). Other studies have shown
similar results with respect to females; however, male estimates differed somewhat (e.g.,
Lynskey and colleagues observed 0.67 for genetic influences in males) (M. T. Lynskey et
al., 2002; Miles et al., 2001).
Although twin studies are able to differentiate between sources of variation due to
genetics and shared and non-shared environments, they do not have the ability to explain
what the actual causal agents are with respect to each influence. Currently, there has
been minimal progress made in identifying the actual genes related to cannabis use;
however, there are some exceptions that will need to be replicated in other samples

14

(Johnson et al., 2008; Stallings et al., 2005; Stallings et al., 2003; Uhl, Liu, Walther,
Hess, & Naiman, 2001). In addition, future studies are needed to elucidate possible
genetic and environmental interplay in relation to cannabis use.

2.1.4.2 Parenting Practices

A number of studies have been conducted with respect to the potential impact of
parent-child relationships and parenting practices on drug use, including cannabis use. A
large body of evidence has linked lower levels of parental monitoring, defined as
tracking, supervision, and knowledge of child whereabouts, with increased risk for drug
use and initiation (Chilcoat & Anthony, 1996; Chilcoat, Dishion, & Anthony, 1995;
DiClemente et al., 2001). Nevertheless, the association has been less well studied in the
context of cannabis initiation and use and the consistency of the findings from the few
studies that have examined the specific link between parental monitoring and cannabis
use is mixed. For example, a cross-sectional study of adolescent females found that teens
with lower parent monitoring had an estimated two-fold increased odds of cannabis use
(DiClemente et al., 2001). Other longitudinal studies have reported similar results;
namely, that low parental monitoring predicted adolescent cannabis use (Hayatbakhsh et
al., 2008; Martins, Storr, Alexandre, & Chilcoat, 2008). With respect to initiation, in a
longitudinal study of Seattle youths followed from the ages of 10 to 18 years, Kosterman
and colleagues (2000) found that parents’ proactive family management, a variable
assessing parents’ monitoring, rules, discipline and reward practices, was associated with
reduced risk of cannabis initiation (Kosterman, Hawkins, Guo, Catalano, & Abbott,

15

2000). A latent class analysis of cannabis use patterns in a sample of black middleschool students also found that lower parental monitoring was associated with increased
odds of membership in a latent class characterized by cannabis use and problems in sixth
grade (Reboussin, Hubbard, & Ialongo, 2007). However, the estimated relationship in
that study was attenuated and non-significant by seventh and eighth grade. It is important
to note that that study focused upon early cannabis involvement and did not cover the
grades or ages when cannabis use is most likely to occur. Similarly, using data from the
National Survey of Parents and Youth, Tang and Orwin (2009) found that parental
monitoring signaled lower odds of cannabis initiation at ages 12 and 13 years, but not for
ages 14 through 16 years (Tang & Orwin, 2009). In other work on parent-child bonding,
authors have suggested that bonding might protect youths and reduce the risk for
cannabis use (Brook, Richter, & Whiteman, 2000).

2.1.4.3 Affiliation with Deviant and Drug Using Peers

Affiliation with drug using peers is one of the strongest and most consistent
predictors of drug initiation in childhood and adolescence (Guo, Hill, Hawkins, Catalano,
& Abbott, 2002; van den Bree & Pickworth, 2005). For example, a longitudinal study of
adolescents in Germany found that higher levels of affiliation with drug using peers
predicted initiation and regular cannabis use (Hofler et al., 1999; von Sydow, Lieb,
Pfister, Hofler, & Wittchen, 2002). These results are similar to findings from
longitudinal studies in the US and New Zealand (Fergusson, Swain-Campbell, &
Horwood, 2002; van den Bree & Pickworth, 2005). In addition, there is some evidence

16

that peers might be more influential in earlier adolescence than later adolescence and
adulthood (Fergusson, Swain-Campbell et al., 2002; Monahan, Steinberg, & Cauffman,
2009); although this finding is not universal (Guo et al., 2002). Moreover, there is a
plausible hypothesized mechanism that links affiliation with drug using peers with
increased opportunities to use cannabis (Lloyd & Anthony, 2003).

2.1.4.4 Intra-Individual Traits and Preexisting Psychiatric Problems

Related personality traits, namely, aggression, delinquency, novelty-seeking, and
risk-taking have been consistently linked with early drug use (Brook, Whiteman, Cohen,
& Tanaka, 1992; Molina & Pelham, 2003; Rios-Bedoya, Wilcox, Piazza, & Anthony,
2008; Rosenberg & Anthony, 2001a; van den Bree & Pickworth, 2005). For example,
Rios-Bedoya et al. (2008) found that a measurement of risk taking was associated with
onset of cannabis use by young adulthood (Rios-Bedoya et al., 2008). Similarly, studies
have reported an association between conduct and/or attention problems and early drug
use, including cannabis use (Fergusson, Horwood, & Ridder, 2007; Molina & Pelham,
2003). In a longitudinal study, Molina and Pelham (2003) found that early adolescent
conduct and attention problems predicted later drug use (Molina & Pelham, 2003).
Another study found a synergistic impact of conduct and attention problems on later drug
use (Flory, Milich, Lynam, Leukefeld, & Clayton, 2003). It is not known, however, if
drug use is an extension of these related traits and earlier behavior problems or a
consequence of them.

17

2.1.5 Mechanisms

2.1.5.1 The Gateway Hypothesis and Exposure Opportunity

A series of reports by Kandel and colleagues have found that cannabis use is often
preceded by tobacco and alcohol use (D. Kandel, 1975; D. Kandel & Faust, 1975; D. B.
Kandel, Yamaguchi, & Chen, 1992; Yamaguchi & Kandel, 1984a, 1984b). Studies have
also shown that regular cannabis users are more likely to go on to use other drugs, such as
cocaine and heroin (D. Kandel, 1975; D. Kandel & Faust, 1975; D. Kandel &
Yamaguchi, 1993). In addition, early cannabis users (i.e., individuals who use before the
age of 18) are more likely to go on to use other drugs (Yamaguchi & Kandel, 1984a,
1984b). These observations led to the development of a “gateway” hypothesis, which
posits that early tobacco and alcohol use leads to cannabis use, which in turn leads to
other drug use, such as cocaine and heroin (D. B. Kandel & Yamaguchi, 1985). It should
be noted, however, that there is considerable debate surrounding the gateway theory (J.
C. Anthony, 2002; Morral, McCaffrey, & Paddock, 2002; Morral, McCafrey, & Paddock,
2002).
The gateway hypothesis is a useful model with respect to relating stages of drug
use; however, it is descriptive in nature and does not explain the underlying mechanisms
responsible for the transition processes. Therefore, several research groups have tested
mechanisms that go beyond the explanatory nature of the gateway process. For example,
in a large national dataset Wagner and Anthony (2002) drew upon the concept of
“exposure opportunity,” i.e., the fact that drug initiation cannot occur unless there is an

18

opportunity to try it, to examine whether individuals who used tobacco and alcohol had
increased opportunities to use cannabis and whether they were more likely to use
cannabis once an opportunity had occurred (Wagner & Anthony, 2002b). The authors
also examined whether cannabis smokers were more likely to use cocaine given an
opportunity to try cocaine (Wagner & Anthony, 2002b). The study found that, compared
with non-users, tobacco and alcohol users were much more likely to have an opportunity
to try cannabis and to use cannabis once an exposure opportunity had occurred (Wagner
& Anthony, 2002b). Cannabis users were also much more likely to use cocaine once an
opportunity had arisen (Wagner & Anthony, 2002b). In addition, the authors concluded
that the observed sequences could not be explained by drug-seeking behavior because of
the time constraints they had placed on the data (Wagner & Anthony, 2002b). In another
test of the association between cannabis use and other subsequent drug use, Fergusson
and Horwood (2000) found that affiliation with drug using peers predicted an increased
risk of drug use; however, it did not explain completely the association between cannabis
use and other drug use (Fergusson & Horwood, 2000). In twin research along these lines,
Gillespie et al. (2009) found that cannabis availability, was the most important factor for
initiation of cannabis smoking (Gillespie, Neale, & Kendler, 2009).
Using a discordant twin design, Lynskey et al. (2003) has probed other aspects of
the gateway hypothesis; namely, the possibility that genetic influences explain the
gateway phenomenon (M. T. Lynskey et al., 2003). The study found that twins who had
used cannabis before the age of 17 years were more likely to use other drugs
subsequently, compared with their co-twins (M. T. Lynskey et al., 2003). The finding
points to the potential importance of other influences, rather than genetic factors, since

19

twins should be no different with respect to other drug use if there was a common genetic
factor (M. T. Lynskey et al., 2003). Other longitudinal and population-based studies have
adjusted for an array of familial and personal factors, including behaviors and attitudes,
and continued to find an association linking prior cannabis use with other subsequent
drug use (Fergusson & Horwood, 2000). Despite the evidence, the possibility remains
that the association between early cannabis use and other drug use is spurious, arising
from a common underlying individual susceptibility.

2.1.5.2 Cannabis Dependence

It is noteworthy that the majority of individuals who initiate cannabis use never
go on to experience the cannabis dependence syndrome (James C. Anthony et al., 1994).
Nonetheless, some users do develop the clinical syndrome of cannabis dependence,
which is likely to be the most common adverse outcome related to cannabis use. The
chief features of cannabis dependence are characterized by disturbances of the mental life
(e.g., recurrent thoughts about use), disturbances of behavior (e.g., repetitions of
cannabis-related behavior), and neuroadaptation related to use (e.g., feeling tolerance) (J.
C. Anthony, 2006).
Using data from the National Comorbidity Survey (NCS), Anthony et al. (1994)
have estimated that one out of every 9 to 11 individuals (~10%) who smoke cannabis
eventually develop the clinical syndrome of cannabis dependence (James C. Anthony et
al., 1994). Wagner and Anthony (2002) have also shown that the estimated risk for
becoming dependent upon cannabis peaks in the second or third year after onset; then,

20

from the fifth year and later drops toward the null (Wagner & Anthony, 2002a). In a
study of the emergence of clinical features among cases of dependence, loss of control
and continuation of use despite harm were the most common initial features (Rosenberg
& Anthony, 2001b).
Cannabis smokers who initiate before adulthood might experience a larger burden
of cannabis dependence. Early onset cannabis smokers have been shown to develop
problems related to their use, i.e., clinical features of dependence, more quickly than
later-onset cannabis users, and are more likely to develop cannabis dependence within 24
months after onset of use, even when elapsed time from onset is taken into account (Chen
et al., 2005). Specifically, the risk might be one in six among those who initiate before
the age of 18 as compared to the overall estimate of one in ten (J. C. Anthony, 2006).
Nonetheless, the causal implications of these findings remain unclear. It is possible that
early-onset cannabis use causes greater risk of becoming dependent; however, individuals
prone to developing dependence might also be more prone to early onset use.

2.1.5.3 Educational Outcomes

A number of studies have found associations linking cannabis use with poor
school performance and reduced educational attainment. A literature review by Lynskey
and Hall (2000) summarized the influence of cannabis use in adolescence on educational
outcomes (M. Lynskey & Hall, 2000). The authors found evidence for the hypothesis
that cannabis use contributes to poor achievement (M. Lynskey & Hall, 2000). They also
suggest the possibility that poor school performance predicts cannabis use, citing

21

additional evidence (M. Lynskey & Hall, 2000). A third possibility is that an underlying
common causal factor influences both cannabis use and poor educational outcomes (e.g.,
externalizing problems); however, empirical examinations of this idea have shown that
even after adjustment for other covariates, cannabis use continues to be associated with
reduced educational attainment (M. Lynskey & Hall, 2000).

2.1.5.4 Other Psychiatric Consequences

A body of evidence links cannabis use, especially early use, with other subsequent
psychiatric complications and co-morbidities. A highly contentious topic is the potential
causal relationship between cannabis use and psychotic symptoms and disorders. For
example, studies on a Swedish male cohort have reported a dose-response relationship
linking the number of times cannabis was used before the age of 18 years and an excess
risk of schizophrenia, even after adjustment for an array of potential confounding
variables such as IQ (Andreasson et al., 1987). These findings have been subsequently
replicated by other studies in other parts of the world (Henquet et al., 2005; van Os et al.,
2002). Self-medication, i.e., cannabis is used with the intention to reduce symptoms of
schizophrenia, has generally not been supported in the literature (Henquet et al., 2005;
van Os et al., 2002). Alternative explanations for the association, namely the possibility
that the association is an artifact of an unmeasured confounder or an underlying
susceptibility for both cannabis use and psychoses, have not been completely ruled out.
A number of studies have also examined the causal relationship between cannabis
use and depression. In comparison to the research on cannabis and schizophrenia,

22

weaker evidence exists for a modest association (O.R. <2.0) between cannabis use and
depression in epidemiologic cross-sectional and longitudinal studies (Chen, Wagner, &
Anthony, 2002; Fergusson, Horwood, & Swain-Campbell, 2002). However, authors of
the studies have not been convinced that they have completely controlled for potential
confounders.
There have been associations linking cannabis use and other psychiatric
symptoms and disorders, as well. For example, in a longitudinal study in New Zealand,
Fergusson and Boden (2008) have reported on the increased risk for adult-onset attention
deficit hyperactivity disorder (ADHD) symptoms in early cannabis users (Fergusson &
Boden, 2008a). However, replication of this finding is warranted. In all of the studies of
psychiatric co-morbidities subsequent to cannabis use a common limitation is the
possibility of uncontrolled background variables or underlying susceptibilities; however,
conducting a study with random assignment to early cannabis use is not possible due to
obvious ethical considerations. Nonetheless, in future research on cannabis cessation
there is the possibility of randomly assigning treatment and observing the occurrence of
secondary endpoints, such as other psychiatric co-morbidities, among the intervention
groups.

2.1.6 Prevention and Control

Epidemiological evidence is scarce with respect to prevention and control of drug
use and/ or cannabis use. A school-based randomized trial on the effectiveness of Drug
Abuse Resistance Education (DARE) showed no school differences with respect to drug

23

use at either five-year or ten-year assessment (Clayton, Cattarello, & Johnstone, 1996;
Lynam et al., 1999). Spoth et al. (2009) have recently reported on a promising schoolbased intervention aimed at reducing opportunities to use drugs and drug use (Spoth,
Guyll, & Shin, 2009). The authors found that the intervention, the Iowa Strengthening
Families Program, was associated with reduced exposure to drug use (odds ratios ranging
from 1.2 to 2.4) and reduced twelfth grade drug use (odds ratios ranging from 2.9 to 6.4)
compared with the control condition (Spoth et al., 2009). Other promising school-based
interventions have been reported, highlighting the need for “booster” follow-up
interventions to improve efficacy (Botvin, Baker, Dusenbury, Botvin, & Diaz, 1995).
Mass media and advertising campaigns aimed at reducing drug use have been less well
studied. Nonetheless, a study by Block et al. (2002) found reduced cannabis use among
individuals able to recall anti-drug advertising (Block, Morwitz, Putsis, & Sen, 2002).
Future research is needed on the effectiveness of population-based intervention and
prevention efforts in all segments of the population.

2.2 Prior Research on Parental Monitoring

The next sections of the dissertation present more information on a plausible
prevention and intervention target and suspected cause of drug use, i.e., parental
monitoring. The definition of parental monitoring and conceptual models linking
parental monitoring with drug use are described in more detail.

24

2.2.1 Definition of Parental Monitoring

Dishion and McMahon (1998) have suggested that parental monitoring, which
includes tracking, supervision, and knowledge of child whereabouts, is one component of
three interrelated dimensions of parenting (Dishion & McMahon, 1998). The other two
dimensions encompass motivation (e.g., norms, values, and goals) and behavior
management (e.g., setting limits and negotiating) (Dishion & McMahon, 1998). Parental
monitoring occurs in the home, school, and community; that is, in all of the environments
of the child (Dishion & McMahon, 1998). It is also a salient part of the child’s
development from infancy to young adulthood and should be contextually and culturally
appropriate (Dishion & McMahon, 1998).

2.2.2 Parental Monitoring Conceptualized within a Developmental Model for Antisocial
Behavior

Patterson et al. (1989) have provided a conceptual framework, the developmental
model for antisocial behavior, which others have called the “social context model”, for
understanding the relationship between parental monitoring and antisocial behavior,
including drug use (Patterson et al., 1989). The model is sketched in Figure 2.1 and
posits that in early childhood ineffective parental practices, including monitoring, lead to
child conduct disorders (Patterson et al., 1989). In turn, the conduct disorders precipitate
academic failure and peer rejection in middle childhood (Patterson et al., 1989). In late
childhood and early adolescence, rejection and academic failure prompt commitment to

25

deviant peer groups and delinquency, including drug use (Patterson et al., 1989). The
model weaves together empirical evidence from previous findings (Patterson et al.,
1989).

26

Figure 2.1 The developmental model for antisocial behavior (Patterson et al. 1989)

27

Recent epidemiological studies have tested components of this model. For
example, Lloyd and Anthony (2003) found that higher levels of parental monitoring were
associated with lower levels of affiliation with deviant peers across late childhood and
early adolescence (Lloyd & Anthony, 2003). Other longitudinal studies have observed
associations linking higher parental monitoring with reduced drug use initiation (Chilcoat
& Anthony, 1996; Chilcoat et al., 1995). However, recent evidence is not consistent and
does not support the link between parental monitoring and poor school performance. For
example, Coley and Hoffman (1996) found that lower levels of monitoring were
associated with higher math achievement scores in third and fourth graders from two
parent families (Coley & Hoffman, 1996). In another study of sixth graders, highest
grade point averages were associated with more moderate levels of parent monitoring
(Kurdek, Fine, & Sinclair, 1995). Therefore, some have re-conceptualized the antisocial
model, which is displayed in Figure 2.2 (Dishion & McMahon, 1998; Lloyd & Anthony,
2003). The model in Figure 2.2 is more similar to recent conceptual models described by
Dishion and McMahon (Dishion & McMahon, 1998). Drawing upon the evidence, this
model links parental monitoring to subsequent affiliation with deviant peers to
subsequent delinquency (e.g., drug use).

28

Figure 2.2 Re-conceptualization of the developmental model for antisocial behavior

29

2.2.3 General Conceptual Models Linking Parental Monitoring with Cannabis Use

The general conceptual model for this dissertation is presented in Figure 2.3. In
this model, there is a direct path from parental monitoring to cannabis use. This model
can be elaborated to test for vectors of possible mediators (e.g. deviant peer affiliation),
vectors of potential markers of subgroup variation or effect-modifiers (e.g. sex and
race/ethnicity), and vectors of other potential explanatory variables. Figure 2.4 depicts
this extension of the general conceptual model.

30

Figure 2.3 The general conceptual model linking parental monitoring with drug use

Parental
monitoring

Drug use

31

Figure 2.4 The general conceptual model linking parental monitoring with drug use extended to include possible mediation, subgroup
variation, and other possible explanatory variables

32

2.3 Conclusions
Cannabis use, especially use before the age of 18 years, is a serious public health
problem. In the US, the treatment burden (in total number of people) associated with
cannabis use is second only to alcohol use (Substance Abuse and Mental Health Service
Administration, 2009). Moreover, early cannabis users have been found to experience
excess problems related to their use, namely DSM-IV drug abuse or dependence, and
have been shown to be more likely to go on to use other drugs (Chen et al., 2005; D.
Kandel, 1975). Recent epidemiological evidence has also suggested a link between early
cannabis use and later adverse health and social outcomes (Brook, Adams, Balka, &
Johnson, 2002). These adverse outcomes include poor educational outcomes and other
psychiatric disturbances, which are associated with decreased productivity and higher
health care costs (Brook et al., 2002; Chen et al., 2005; Fergusson & Boden, 2008a; M.
Lynskey & Hall, 2000; Tien & Anthony, 1990; Wagner & Anthony, 2002a, 2002b).
Hence, there is a substantial need for the identification of possible targets for prevention
and early intervention of cannabis use. This dissertation research will use data from two
longitudinal studies to probe the long-term influence of a suspected causal agent and
potential target for cannabis prevention, parental monitoring.

33

3. CHAPTER THREE - MANUSCRIPT ONE

PARENTAL MONITORING AT AGE 11 AND SUBSEQUENT ONSET OF
CANNABIS USE UP TO AGE 17: RESULTS FROM A PROSPECTIVE STUDY

ABSTRACT

Background: Early-onset cannabis smoking is a risk marker for subsequent adverse
psychiatric outcomes. Delay or prevention of early-onset drug use might be achieved via
parenting interventions such as programs to increase parents’ effective monitoring and
supervision of their children. The aim of this study is to estimate, prospectively, the
influence of parental monitoring assessed at age 11 on the initiation of cannabis use
before age 18 years.
Methods: Data are from a longitudinal study of 823 low birth weight and normal birth
weight children randomly selected from 1983-1985 newborn discharge lists of two major
hospitals in southeast Michigan, one serving inner-city mothers and the other, suburban
mothers. Parental monitoring was assessed at age 11 via 10 items, and the parental
monitoring – cannabis prediction was estimated for the 641 children who were assessed
at baseline, at age 11 years, and at age 17 years. Poisson regression with robust error
variances was used to estimate the predictive association that links levels of parental
monitoring at age 11 with the risk of cannabis use up to age 17, adjusting for other
important covariates.

34

Results: Higher levels of parental monitoring at age 11 were associated with a reduced
risk of cannabis initiation from ages 11 to 17 years (adjusted estimated relative risk =
0.96; 95% confidence interval = 0.94, 0.98).
Conclusions: This prospective investigation found that higher levels of parental
monitoring predicted a reduced occurrence of cannabis initiation from ages 11 to 17
years. Consistent with evidence reported elsewhere, these findings from prospective
research help confirm a theory that parenting and familial characteristics might exert
long-lasting influences on a child’s risk of starting to use illegal drugs.

35

3.1 Introduction

In the United States in 2008 more than two million individuals aged 12 years and
older smoked cannabis for the first time (Substance Abuse and Mental Health Service
Administration, 2009). Most of these recent initiates were adolescents younger than 18
years of age. This group of early-onset cannabis users (i.e., those who started to use
before the age of 18) is of particular public health salience for several reasons. Early
onset cannabis smokers are more likely to develop cannabis dependence within 24
months after onset of use, even when elapsed time from onset is taken into account (Chen
et al., 2005). It is important to note, however, that the causal implications of these
findings remain unclear. The possibility of a common underlying susceptibility for both
early cannabis use and cannabis dependence has not been completely ruled out. Early
cannabis users are also more likely to go on to use other drugs (D. B. Kandel, 1984;
Yamaguchi & Kandel, 1984a), possibly due to increased opportunities to try these other
drugs (Wagner & Anthony, 2002b). In addition, there is some indication that cannabis
users might be more likely to experience other subsequent psychiatric problems
(Fergusson & Boden, 2008a, 2008b; Tien & Anthony, 1990).
Given these findings, early adolescence might be a critical developmental interval
with respect to preventing or delaying onset of cannabis use. Parental monitoring (i.e.,
awareness, tracking, and supervision of children’s activities) is a specific facet of
parenting influence that may lend itself to preventive approaches. A series of studies
have estimated associations linking higher levels of parental monitoring with lower odds
of drug initiation (Chilcoat & Anthony, 1996; Chilcoat et al., 1995; DiClemente et al.,
2001). Several potential mechanisms have been posited. One suggestion is that children
36

with higher levels of parental monitoring are less likely to associate with deviant peers,
thereby reducing their exposure to drugs (Lloyd & Anthony, 2003). In twin research
along these lines, Gillespie et al. (2009) found that cannabis availability, which the
authors noted might be determined by aspects of parent-child relationships like
monitoring, was the most important factor for initiation of cannabis smoking (Gillespie et
al., 2009).
Against this background of empirical evidence and a well-reasoned theoretical
framework to link parental monitoring with later drug use initiation in children and
adolescents (Dishion & McMahon, 1998), several gaps in the literature remain. First,
few epidemiologic studies have specifically examined cannabis smoking in the past
research on the parental monitoring-drug initiation relationship. Most prior research has
grouped all drugs together or has only examined tobacco or alcohol initiation. However,
as described above, early-onset cannabis smokers represent a potentially important risk
subgroup. Second, previous studies on parent monitoring and cannabis initiation have
not always considered potentially important covariates, namely, maternal smoking and
peer influences. Third, it is uncertain if the association between parental monitoring and
drug use is uniform across certain subgroups of the population, specifically across
subgroups defined by sex and race/ethnicity. For example, a previous study found that
higher levels of parental monitoring were associated with lower odds of tobacco smoking
initiation only among white adolescents (Bohnert, Rios-Bedoya, & Breslau, 2009).
Fourth, previous studies have had relatively short follow-up periods, usually no more
than two years, and have not covered the ages when cannabis initiation is most likely to
occur.

37

Here, in this new research, the focus is on early-onset cannabis smoking, and the
epidemiological evidence is from a cohort of urban and suburban children from a large
Midwest metropolitan area. The aim is to estimate prospectively the relationship that
links parental monitoring in pre-adolescence (age 11 years) with the cumulative
incidence of cannabis use up to late adolescence (age 17 years). The present
investigation overcomes the gaps in the literature in three specific ways. First, the latechildhood and adolescent intervals that are covered are the times when cannabis initiation
is most likely to occur and where problems related to use appear most frequently (i.e., the
clinical features of cannabis dependence and the clinical syndrome of cannabis
dependence). Second, tests are conducted to examine whether the association between
monitoring and onset of cannabis use varies between blacks and whites, and for males
versus females. Third, the investigation takes into account important covariates,
including peer smoking and drinking, as well as maternal smoking.

3.2 Methods

3.2.1 Sample

Data are from a longitudinal study on the neuropsychiatric sequelae of low birth
weight (LBW) and normal birth weight (NBW) children. Detailed information on the
sample is available elsewhere (Breslau et al., 1996), and is briefly summarized here.
LBW and NBW children were randomly selected from 1983-1985 newborn discharge
lists of two hospitals in southeast Michigan, one serving an inner-city community and the

38

other serving a suburban community. Forty-seven children with severe neurologic
impairments were excluded from the initial sample. Of the 1095 children eligible for the
study, 823 (75.2%) participated in the initial assessment from 1990 to 1992, when they
were six years of age. Follow-up assessments were conducted at 11 years (n=717;
87.1%) and 17 years of age (n=713; 86.6%). Six hundred and fifty-seven children
completed both age 11 and age 17 follow-up assessments. The institutional review boards
of the participating institutions approved the study.

3.2.2 Measures

3.2.2.1 Cannabis Initiation, Ages 11-17 Years

The cumulative incidence of cannabis use up to age 17 was assessed during the
age 17 assessment via the following dichotomous (coded Yes (1) or No (0)) question
about cannabis use: “Have you ever, even once, used marijuana or hashish?” Data from
the age 11 assessment were used to identify children who had initiated cannabis smoking
by age 11 and were therefore no longer at risk for initiation during the 11 to 17 years agespan. In addition, for those respondents who had initiated cannabis by age 17, a followup question queried the age of their first cannabis use.

39

3.2.2.2 Parental Monitoring, Age 11 Years

Level of parental monitoring was assessed by child self-report at age 11 via a
standardized 10-item scale (the full scale is attached in Appendix D) (Capaldi &
Patterson, 1989; Chilcoat et al., 1995). The items encompass child supervision and
tracking of activities outside the school environment (e.g., whether an adult was present
within 1 hour of the child arriving home from school, how often the child talked with the
parents about plans for the coming day, and whether the child knew how to contact the
parents if they were not at home after school). On seven of the items, children responded
on a five-point scale ranging from All of the time (1) to Never (5); responses were coded
either Clear (1) or Unclear (2) on two items; and on a single item the responses were
coded Yes (1) or No (2). A parental monitoring score was constructed by reversing the
coding and summing the scores on the 10 items. Possible scores ranged from a low of 10
to a high of 41.

3.2.2.3 Child Covariates

Tobacco smoking and alcohol use at age 11 were assessed via standardized childreported drug questions at age 11 years. Peer smoking and alcohol use was assessed by
child self-report at age 11 years via two Yes or No questions: “Do you have any friends
around your age who ever smoke tobacco cigarettes?” and “Do you have any friends
around your age who ever drink alcohol?” Community, Sex, and Race were as assessed
at baseline. Birth weight was obtained from hospital records.

40

3.2.2.4 Maternal Covariates

Maternal smoking was assessed at baseline when children were 6 years of age.
Mothers were classified as smokers if they had ever smoked daily for one month or more
up to the time of the interview. Maternal education and maternal marital status were also
assessed at baseline.

3.2.3 Statistical Analysis

Using a method described by Zou (2004), unadjusted and adjusted Poisson
regressions with robust error variances were conducted to estimate the predictive
association linking parental monitoring at age 11 years with the initiation of cannabis use
from the ages of 11 to 17 years (Zou, 2004). This estimate has been shown to yield more
precise estimates of relative risk than the odds ratio derived in traditional logistic
regression, especially when the outcome of interest occurs in greater than 10% of the
sample (i.e., when estimated odds ratios may not approximate relative risks) (Zou, 2004).
Male-female subgroup variation in the parental monitoring-cannabis initiation association
was evaluated via product terms, as was subgroup variation associated with race. None
were detected at the alpha level of 0.05.
For a subsidiary survival analysis, a dichotomous variable was constructed from
the original parental monitoring scale using the median as the cutoff (low parental
monitoring (<37) and high parental monitoring (≥37)). Next, Kaplan-Meier curves were
derived for the above- and below-median monitoring groups in order to inspect the

41

failure rates of first cannabis use, year by year from age 11 to 17 years. Specifically, the
approach involved specifying the elapsed time from the age at the age 11 assessment until
the age of first cannabis use. Adolescents who never used cannabis contributed personyears up to the time of their age at the age 17 interview. A total of 619 children with age
of onset information contributed data to the Kaplan-Meier analysis. A log-rank test was
conducted to formally test whether the two survival curves differed from one another
(alpha = 0.05).
Finally, in a post-estimation exploratory step, the method of plotting fractional
polynomials, as described by Royston and Altman (1994), was employed to probe into
the issue of possible non-linearity in the parental monitoring – cannabis initiation
association (Royston & Altman, 1994). All analyses were conducted using Stata 11
(StataCorp, 2009).

3.3 Results

3.3.1. Comparison of the Original Sample with the Follow-up Sub-samples at Ages 11
and 17 Years

Six hundred fifty seven of the initial 823 children participated in both waves of
follow-up assessment. For focus on cannabis onset after level of parental monitoring was
assessed, individuals who had used cannabis before age 11 years were excluded (n=7).
Nine children had missing information on one or more of the covariates of interest.
Therefore, the resulting sample size for the analysis was n=641. As shown in Table 3.1,

42

the initial sample differs negligibly from the subset followed up at age 11 and 17 years
and from the subset used in the regression analyses, after exclusions noted above.

43

Table 3.1 Description of the initial sample, the subset with
follow-up data for ages 11 and 17, and the subset used in the
regression analyses. Data come from 823 children sampled from
1983-1985 newborn discharge lists in southeast Michigan and
assessed at ages 6, 11, and 17 years.
Sample with Sample used
follow-up
in the
Initial
data for ages
regression
sample
11 and 17
analyses
(n=823)
(n=657)
(n=641)
%
%
%
Urban
50.2
52.4
51.8
Low birth weight
57.5
56.9
56.8
Black
42.9
46.3
45.4
Male
48.6
46.6
46.2
Mothers’ education
< High school
16.9
16.0
15.6
High school
27.5
26.2
26.5
Some college
37.3
38.5
38.4
College
18.3
19.3
19.5
Single mother
32.8
32.6
32.4

44

3.3.2 Parental Monitoring at Age 11 and the Risk of Cannabis Initiation up to Age 17

An estimated 35% of the 641 individuals started smoking cannabis during the
follow-up period, from ages 11 to 17 years. The third column in Table 3.2 depicts the
cannabis initiation rate during follow-up (%) for each of the covariates of interest. Table
3.2 also depicts estimates from the unadjusted Poisson regressions linking the covariates
of interest with the cumulative incidence of cannabis use from age 11 up to age 17 years;
Table 3.3 depicts the adjusted regressions. With respect to parental monitoring, an
increase of one point on the parental monitoring scale signaled an estimated five percent
decrease in the likelihood of initiating cannabis smoking by age 17 years (estimated
relative risk, ERR, = 0.95; 95% confidence interval, CI, = 0.93, 0.97). A slightly
attenuated but statistically robust association remained after adjustment for other
important covariates (adjusted ERR, AERR, = 0.96; 95% CI = 0.94, 0.98). No subgroup
variation for sex or race was detected at the alpha level of 0.05 (i.e., product-terms for sex
and race with parental monitoring were not statistically significant).
Other findings of interest are presented in the adjusted table (Table 3.3).
Compared with white adolescents, black adolescents were less likely to initiate cannabis
use from age 11 to age 17 years (AERR = 0.64; 95% CI = 0.48, 0.87). Males were more
likely to initiate cannabis use during follow-up (AERR = 1.25; 95% CI = 1.01, 1.56).
Smoking tobacco by age 11 was associated with an increased risk of cannabis initiation
from the ages of 11 to 17 years (AERR = 1.31; 95% CI = 1.02, 1.69). Having a friend
who smoked at age 11 was also robustly associated with initiating cannabis during
follow-up (AERR = 1.63; 95% CI = 1.31, 2.03). With respect to maternal covariates,

45

having a single mother at baseline and having a mother who smoked tobacco at baseline
were associated with an increased risk of cannabis initiation during follow-up (AERR =
1.63; 95% CI = 1.27, 2.09, and AERR = 1.54; 95% CI = 1.25, 1.91, respectively).

46

Table 3.2 Parent monitoring at age 11 and initiation of cannabis use up to age 17,
unadjusted models. Data come from 641 children sampled from 1983-1985 newborn
discharge lists in southeast Michigan with complete information on all variables.
% cannabis Estimated
95%
initiation 11 Relative Confidence
n
to 17 years
Risk
Interval
p value
Parent Monitoring
641
34.6
0.95
(0.93, 0.97) <0.001
Birth weight
Low
364
33.8
0.95
(0.76, 1.17)
0.607
Normal
277
35.7
Ref
Community
Urban
332
34.3
0.98
(0.79, 1.22)
0.870
Suburban
309
35.0
Ref
Race
Black
291
32.3
0.88
(0.71, 1.10)
0.261
White
350
36.6
Ref
Sex
Male
296
41.6
1.45
(1.17, 1.79)
0.001
Female
345
28.7
Ref
Smoked Tobacco by age 11
Yes
64
62.5
1.98
(1.58, 2.48) <0.001
No
577
31.5
Ref
Drank alcohol by age 11
Yes
64
53.1
1.63
(1.26, 2.11)
0.001
No
577
32.6
Ref
Had a friend who smoked
tobacco at age 11
Yes
180
52.8
1.92
(1.56, 2.35) <0.001
No
461
27.5
Ref
Had a friend who drank
alcohol at age 11
Yes
516
49.6
1.60
(1.29, 1.99) <0.001
No
125
31.0
Ref
Maternal Education
< High School
100
37.0
Ref
High School
170
37.1
1.00
(0.73, 1.38)
0.992
Some College
246
33.7
0.91
(0.67, 1.24)
0.560
College
125
31.2
0.84
(0.59, 1.21)
0.360
Maternal Marital Status
Single
208
42.3
1.37
(1.11, 1.69)
0.004
Married
433
30.9
Ref
Maternal tobacco smoking
at baseline
Smoker
234
45.7
1.62
(1.31, 1.99) <0.001
Non-smoker
407
28.3
Ref

47

Table 3.3 Parent monitoring at age 11 and initiation of cannabis use up to age 17,
adjusted model. Data come from 641 children sampled from 1983-1985 newborn
discharge lists in southeast Michigan with complete information on all variables.
Adjusted
Estimated
Relative
95% Confidence
Risk
Interval
p value
Parent Monitoring
0.96
0.94, 0.98
<0.001
Birth weight
Low
0.85
(0.69, 1.04)
0.113
Normal
Ref
Community
Urban
1.01
(0.75, 1.37)
0.925
Suburban
Ref
Race
Black
0.64
(0.48, 0.87)
0.004
White
Ref
Sex
Male
1.25
(1.01, 1.56)
0.042
Female
Ref
Smoked Tobacco by age 11
Yes
1.31
(1.02, 1.69)
0.035
No
Ref
Drank alcohol by age 11
Yes
0.98
(0.73, 1.33)
0.899
No
Ref
Had a friend who smoked tobacco
at age 11
Yes
1.63
(1.31, 2.03)
<0.001
No
Ref
Had a friend who drank alcohol at
age 11
Yes
1.25
(0.99, 1.57)
0.061
No
Ref
Maternal Education
< High School
Ref
High School
1.17
(0.85, 1.59)
0.338
Some College
1.06
(0.79, 1.44)
0.687
College
1.33
(0.91, 1.95)
0.144
Maternal Marital Status
Single
1.63
(1.27, 2.09)
<0.001
Married
Ref
Maternal tobacco smoking at
baseline
Smoker
1.54
(1.25, 1.91)
<0.001
Non-smoker
Ref
48

3.3.3 Kaplan-Meier Survival Estimates

Figure 3.1 depicts estimates from the Kaplan-Meier analysis. The estimated
cumulative incidence of cannabis use by age is displayed for the above- and belowmedian monitoring groups. As shown in the figure, the failure estimates for the lower
parental monitoring group were consistently higher than those for the higher parental
monitoring group. That is, by the end of each age interval adolescents in the lower
parental monitoring group were more likely to have initiated cannabis smoking. By age
17, an estimated 45% of the individuals in the lower monitoring group had initiated
cannabis use, compared with an estimated 28% in the higher monitoring group. The logrank test revealed that the curves for the two parental monitoring groups were
significantly different from one another (p<0.001).

49

Estimated cumulative incidence of cannabis use
0.00
0.25
0.50
0.75
1.00

Figure 3.1 Kaplan Meier estimates of cumulative incidence of cannabis use from age 11 to 17 years. Data come from 619 children
sampled from 1983-1985 newborn discharge lists from two hospitals in southeast Michigan.

11

13

Age (years)

High parental monitoring

15
Low parental monitoring

50

17

3.3.4 Fractional Polynomial Post-estimation Examination of the Parental Monitoring –
Cannabis Initiation Association

In post-estimation analyses, the method of plotting fractional polynomials was
used to probe into the issue of non-linearity in the estimated association that links early
parental monitoring with later cannabis initiation. Figure 3.2 depicts the fractional
polynomial plot of the estimated probability of cannabis initiation from 11 to 17 years
with 95% confidence intervals across levels of parental monitoring. With the exception
of the very low end of the monitoring distribution, where sparse data created wide
confidence bounds, there is a consistent linear decline in the probability of cannabis
initiation for increasing scores on the parental monitoring scale.

51

Figure 3.2 Estimated probability of cannabis initiation from ages 11 to 17 years by level of parental monitoring. Data come from 641
children sampled from 1983-1985 newborn discharge lists in southeast Michigan with complete information on all variables.

52

3.4 Discussion

In this longitudinal study of a hypothesized predictive relationship linking
parental monitoring levels at age 11 years with the cumulative incidence of cannabis use
up to age 17, a primary finding was that children with higher levels of parental
monitoring were less likely to initiate cannabis use across the age-span from 11 to 17
years. Specifically, for every one unit increase on the parent monitoring scale, there was
an estimated four percent reduction in the likelihood of cannabis initiation, with statistical
adjustment for other covariates. Children in the lower parental monitoring group at age
11 were more likely than children in the higher monitoring group to initiate cannabis use
at each age interval from 11 to 17 years. Estimates from the fractional polynomial
analysis suggest no attenuation of the relationship at the highest levels of parental
monitoring. That is, parents at the highest parental monitoring level at age 11 years did
not seem to induce a negative reaction and “acting out” in response to these higher
parental monitoring levels.
Several strengths and selected limitations should be considered before a more
detailed discussion of the findings. A major strength is the epidemiological frame for the
original study sample. This was not a sample of delinquent or drug-involved youths; nor
was it a sample of high-risk youths as often has been the case in research on family
factors and parenting in relation to adolescent-onset drug use. Furthermore, the
assessments of levels of parental monitoring were taken more than five years before the
assessment of the cannabis initiation outcomes. This strength of the research design
reduces a threat to validity of the study estimates in the form of inadvertent reciprocity

53

(i.e., the possibility that a child’s expression of curiosity or interest in cannabis smoking
might cause parents to increase their monitoring and supervision levels). It also reduces
the threat of “shared methods covariation,” as might be induced when the assessments of
monitoring and outcomes are completed after a short-span follow-up (e.g., one year
apart).
There are also several potential limitations to note. First, although the findings
indicate a robust and stable association between early monitoring and the risk of cannabis
initiation later on, an important question remains. Does early parental monitoring have a
lasting influence regardless of its continuity into late adolescence? The apparent long
term benefits of early parent monitoring might reflect stability of parenting behavior over
time, as children mature, rather than an investment that pays off later on by deterring
adolescents from involvement with cannabis. Future studies might be more illuminating
if they included re-assessments of parental monitoring during follow-up intervals to test
this question. Second, it is possible that sample attrition influenced the findings.
However, there were no differences on sample characteristics between the initial and
follow-up samples to support this assertion. Third, due to the observational nature of this
study and the assumption that no other potential explanatory variables were omitted,
causal interpretations are not warranted at this time. Replications and studies that test the
outcomes of interventions directed toward enhancing parental monitoring are needed.
The findings from the present study, focused on cannabis initiation in
adolescence, are consistent with a large body of evidence linking lower levels of parental
monitoring with increased odds of drug use (Chilcoat & Anthony, 1996; Chilcoat et al.,
1995; DiClemente et al., 2001). With respect to the specific hypothesized link between

54

parental monitoring and cannabis use, the findings from the present investigation
generally converge with the findings of the few published studies; however, there are
some differences. For example, the findings from the present investigation are consistent
with a cross-sectional study of adolescent females that found that teens with lower parent
monitoring had an estimated two-fold increased odds of cannabis use (DiClemente et al.,
2001). Similarly, in a longitudinal study of Seattle youths followed from the ages of 10
to 18 years, Kosterman and colleagues (2000) found that parents’ proactive family
management, a variable assessing parents’ monitoring, rules, discipline and reward
practices, was associated with reduced risk of cannabis initiation (Kosterman et al.,
2000). A latent class analysis of cannabis use patterns in a sample of predominantly
black middle-school students found that lower parental monitoring was associated with
increased odds of membership in a latent class characterized by cannabis use and
problems in sixth grade (Reboussin et al., 2007). However, this relationship was
attenuated and non-significant by seventh and eighth grade. It is noteworthy that
Reboussin and colleagues (2007) examined early cannabis involvement and did not cover
the grades or ages when cannabis use is most likely to occur during adolescence.
Similarly, using data from the National Survey of Parents and Youth, Tang and Orwin
(2009) found parental monitoring signaled lower odds of cannabis initiation at ages 12
and 13 years, but not for ages 14 through 16 years (Tang & Orwin, 2009). Both studies
suggested that other factors, namely peers, might have become more influential than
parents in later years. In contrast, in the present study both parental monitoring and
affiliation with tobacco using peers were important predictors of cannabis initiation, and

55

that the influence of parental monitoring on cannabis initiation was stable over the ages
from 11 to 17 years.
In a previous report on this follow-up study the influence of parental monitoring
on tobacco initiation varied with respect to race/ethnicity (Bohnert et al., 2009). Lower
levels of parental monitoring were linked with higher odds of tobacco initiation only
among white adolescents. In the present investigation, subgroup differences with respect
to race/ethnicity on the parental monitoring - cannabis initiation relationship were not
detected. The reasons for these differences between the two investigations are unclear,
but reduced statistical power to detect sub-group variation might be at play. In both
studies black adolescents had lower levels drug initiation; this finding is consistent with a
number of empirical studies and results from national school-based surveys (Guo et al.,
2002; Johnston et al., 2009).
Affiliation with drug using peers is one of the strongest and most consistent
predictors of drug initiation in childhood and adolescence (Guo et al., 2002; van den Bree
& Pickworth, 2005). Some have suggested that peers might be more influential than
parents or more influential at older ages than parents with respect to cannabis smoking
(Guo et al., 2002; Reboussin et al., 2007; Tang & Orwin, 2009; van den Bree &
Pickworth, 2005). In the present study, early affiliation with peers who used tobacco was
related to cannabis initiation. However, the inclusion of affiliation with tobacco and
alcohol using peers as covariates did not appreciably dampen the association of parental
monitoring with cannabis initiation in the adjusted model. Both were important
predictors of cannabis initiation.

56

Consistent with evidence from prior studies, the present findings, from
prospective research, help confirm one conceptual hypothesis, i.e., that parenting and
familial characteristics might exert long-lasting influences on a child’s risk of starting to
use illegal drugs. The present study focused on the interval from late childhood to the
end of adolescence, covering ages when cannabis initiation in the Unites States is most
likely to occur. Trials aimed at strengthening parenting practices are underway and some
of them show promise (Spoth et al., 2009). Future prevention research should continue to
evaluate the effect of interventions aimed at promoting parent monitoring. Additional
longitudinal studies will be necessary to help elucidate the possible paths from parental
monitoring to drug use.

57

4. CHAPTER FOUR - MANUSCRIPT TWO

LONGITUDINAL STUDY OF LEVELS OF ADOLESCENT DRUG USE: AN SEM
ANALYSIS OF PARENTAL MONITORING

ABSTRACT

Aims: To estimate the influence of level of parental monitoring (PM) assessed at age 11
on level of drug use at age 17, simultaneously testing paths through levels of drug use
and affiliation with drug using peers (DUP) at age 11.
Setting: Data are from children born in 1983-1985, sampled and recruited in southeast
Michigan.
Participants: Of the original cohort (n=823) children, 774 (87%) contributed data to the
present study.
Measurements: Standardized interviews assessed drug use at ages 11 and 17, and PM and
DUP at age 11. A structural equation model (SEM) was used to test paths of PM with
DUP and drug use at ages 11 and 17.
Findings: According to the pre-specified SEM, the estimated level of drug use at age 11,
as well as DUP, depended upon PM level, which also predicted drug use at age 17. DUP
at age 11 did not predict drug use at age 17.
Conclusions: Level of PM may influence levels of drug use over a longer time span than
previously believed (six years). This evidence helps substantiate the theory that
parenting characteristics might exert long-lasting influences on a child’s use of drugs.

58

4.1 Introduction

Alcohol, tobacco, and cannabis are three of the most widely used psychoactive
drugs in the world (Degenhardt et al., 2008). In the United States, experimentation and
use of these drugs is common, often occurring before the age of 18 years (Substance
Abuse and Mental Health Service Administration, 2009). Drug use before adulthood,
however, is associated with adverse psychiatric and social outcomes. For example, earlyonset drug users might be more likely to experience problems related to their use,
namely, DSM-IV drug abuse or dependence, compared with those who start using later in
life (Chen et al., 2005; Grant et al., 2006; Winters & Lee, 2008). This association is not
entirely explained by having greater elapsed time from onset of first drug use (Chen et
al., 2005). Nonetheless, early-onset use might not be causally related to later drug
problems, i.e., individuals prone to early use might also be more liable to experience drug
problems. There is also evidence that links early-onset use with an excess risk for other
later psychiatric disturbances (de Graaf et al.; Fergusson & Boden, 2008a; Tien &
Anthony, 1990) and to poorer school performance and lower educational attainment (M.
Lynskey & Hall, 2000); although the causal significance of these associations remains
uncertain.
It follows that prevention or delay of the onset and use of alcohol, tobacco, and
cannabis has salience in public health work. Most of the public health work along these
lines has a focus on school and/or family. For example, Botvin and colleagues have
conducted large-scale randomized trials in school-based settings (Botvin et al., 2000). In
contrast, interventions developed on the basis of the social learning theory as applied to
the family often are directed toward high risk family environments (e.g., (Kumpfer &
59

Alvarado, 2003)) or enhancement of positive parenting practices (e.g., (Dishion &
Kavanagh, 2000)), such as the surveillance and monitoring conducted by parents during
the childhood and adolescent years (i.e., parental monitoring (PM)). Implicit in these
school and family oriented interventions is an assumed importance of affiliation with
drug using or otherwise rule-breaking peers (i.e., deviant peers).
An important issue that has surfaced in the school intervention research involves
the equivalent of “booster shots” in public health vaccination initiatives. Namely, it
seems that many early primary prevention intervention effects fade as children mature
through the school years. This fading of early intervention effects has motivated a line of
research on “booster” drug prevention programs (Botvin et al., 1995). The issue of
intervention program “fade” and the need for “booster” programs has been less well
studied in the context of family centered studies, perhaps based upon an assumption that
the effects of PM and other family-level parenting practices are inherently transient, and
less important as the child progresses into middle and late adolescence.
An alternative to the “transient effects” assumption about PM and other positive
parenting effects is a “persistent effects” hypothesis. A test of the “persistent effects”
hypothesis requires: (a) evidence of an inverse association that links levels of PM with
levels of drug use during the childhood years, and (b) evidence of an inverse association
that links early PM levels with later levels of drug use (e.g., use in middle to late
adolescence). In most of the prior research on PM, the assumption of “transient effects”
has been made (e.g., Lloyd and Anthony, 2003). Here, the focus of inquiry is to test
these alternative specifications for “transient” versus “persistent” effects of PM, as a
guide toward future enhancement of parenting programs to prevent youthful drug use.

60

Clearly, if the “transient effect” assumption holds, then booster sessions for PM might be
required. Of course, the idea of “booster” programs to help parents maintain adept levels
of PM has some complex facets, in addition to logistical problems faced when
interventionists have attempted to bring parents in for group prevention programs or to go
out to parents in their homes.
To a lesser extent, these same kinds of processes might be taking place in the
context of peer influences on drug use. Evidence is conflicting with respect to whether
peer influence is a relatively short-lived phenomenon in childhood and adolescence (i.e.,
transient)(e.g., (Steinberg & Monahan, 2007)) or more persistent (e.g., (Guo et al.,
2002)).
Hence, the aim of the present study is to use a representative cohort to evaluate
the potential persistent influence of parental monitoring assessed at age 11 on levels of
drug use at age 17, while simultaneously examining potential paths through levels of drug
use and affiliation with drug using peers at age 11 years.

4.2 Methods

4.2.1 Data and Sample

Longitudinal data come from a 1983-1985 birth cohort of children from southeast
Michigan. Complete information on the population, sampling procedures, and
assessments is available elsewhere (Breslau et al., 1996), and is briefly summarized here.
Low birth weight and normal birth weight children were randomly selected from 1983-

61

1985 newborn discharge lists of two hospitals in southeast Michigan, one serving a
disadvantaged urban community and the other serving a middle-class suburban
community. At the time of recruitment, 47 children with severe neurologic impairment
were excluded from the initial sample. A total of 823 (75.2%) out of the 1095 children
eligible for the study participated in the initial assessment from 1990 to 1992, when they
were six years of age. Follow-up assessments were conducted when the children were 11
(n=717; 87.1%) and 17 years of age (n=713; 86.6%). Children were assessed via
standardized face-to-face interviews. Of the original cohort of 823 children, 774 (94.0%)
contributed data to the present investigation. The institutional review boards of the
participating institutions approved the study.

4.2.2 Measures

4.2.2.1 Parental Monitoring

Parental monitoring (PM 11) was elicited from the children when they were 11
years of age using a 10-item scale (Appendix D) (Capaldi & Patterson, 1989; Chilcoat et
al., 1995). The scale encompasses information on parental supervision and tracking of
the child. A summary score was generated by reversing the coding and summing the
scores. Scores ranged from a low of 10 to a high of 41, with higher scores indicating
higher levels of monitoring.

62

4.2.2.2 Drug Use

Level of drug use at age 17 (Drug 17) was constructed as a continuous latent
variable from the following three dichotomous (yes/no) child-reported questions on
alcohol, tobacco cigarette smoking, and cannabis smoking, respectively:
•

“Have you ever, even once, had a drink of any type of alcoholic beverage?
Do not include sips from another person’s drink.”

•

“Have you ever smoked a cigarette, even one or two puffs?”

•

“Have you ever, even once, used marijuana or hashish?”

Level of drug use at age 11 (Drug 11) was constructed as a continuous latent
variable from the following child-reported age of onset questions which were coded into
dichotomous used/never used variables for alcohol, tobacco cigarette smoking and
cannabis smoking, respectively:
•

“How old were you the first time you drank beer, wine, wine coolers,
liquor, or any other drink with alcohol in it?”

•

“How old were you when you first smoked a tobacco cigarette, even just a
puff?”

•

“How old were you the first time you smoked (marijuana/reefer)?”

63

4.2.2.3 Affiliation with Drug Using Peers

Level of affiliation with drug using peers at age 11 (DUP 11) was constructed as a
continuous latent variable from the following three dichotomous (yes/no) child-reported
questions on peer alcohol, tobacco cigarette smoking, and cannabis smoking,
respectively:
•

“Do you have any friends around your age who ever drink alcohol?”

•

“Do you have any friends around your age who ever smoke tobacco
cigarettes?”

•

“Do you have any friends around your age who ever smoke
(marijuana/reefer)?”

4.2.2.4 Baseline Covariates

The two stratification variables, birth weight and community, as well as sex, and
maternal education were measured at baseline.

4.2.3 Statistical Analysis

Structural equation modeling (SEM) was used to simultaneously test paths of
PM11 with DUP 11 and Drug 11 and 17. The final model was adjusted for the two
stratification variables, birth weight and community, as well as, sex and level of maternal

64

education. All models were fit in Mplus 6.0 using the weighted least squares means and
variance adjusted estimator (Muthén & Muthén, 1998-2010). The latent variable SEM
approach provides several advantages over traditional regression techniques. First, it
allows for simultaneous testing of multiple regression paths of complex relationships.
Second, common sources of variation between outcome variables are taken into account
and the path estimates are unencumbered by assumptions about error. Third, Mplus
handles missing data using multiple imputation methods to use all available data (Muthén
& Muthén, 1998-2010). Model fit was assessed via the Comparative Fit Index (CFI)
(where good model fit is denoted by a score >0.90) (Bentler, 1990), the Tucker-Lewis
Index (TLI) (where good model fit is denoted by a score >0.90)(Bentler & Bonett, 1980),
and the Root Mean Square Error of Approximation (RMSEA) (where good model fit is
denoted by a score <0.05) (Bentler, 1990).

4.3 Results

4.3.1 Comparison of the Original Sample with the Subset Used in the Analysis

Forty-nine individuals were excluded from the analytic sample because they did
not have data on any of the outcomes in the structural equations model. Table 4.1
presents a comparison of the original sample at age 6 years (n=823) and the analytic
sample used in the SEM (n=774). The two samples did not differ on any key sample
characteristics (p>0.05).

65

Table 4.1 Description of the initial sample and the subset used in the structural
equation model (SEM). Data come from 823 children sampled from 1983-1985
newborn discharge lists in southeast Michigan and assessed at ages 6, 11, and
17 years.
Initial sample
Sample used in the
(n=823)
SEM
%
(n=774)
%
Low birth weight
57.5
57.2
Urban
50.2
50.9
Male
48.6
48.1
Maternal Education
Less Than High School
16.9
16.7
High School Graduate
27.5
26.9
Some College
37.3
37.6
College Graduate
18.4
18.9

66

4.3.2 Mean Levels of Parental Monitoring and Percentages of Drug Use in the Sample

As shown in Table 4.2, the mean score for parental monitoring was 36.1 (standard
deviation = 4.0). With respect to affiliation with drug using peers, almost 30% of the
children had a friend who smoked tobacco cigarettes at age 11 years, the highest
percentage of affiliation among the three drugs (i.e., percentages of affiliation with
alcohol and cannabis using peers were lower). Table 2 also depicts percentages of drug
use at ages 11 and 17 years. The proportion of children using drugs at age 11 was
relatively low. The percentage with a history of alcohol use was about 10%; nearly
identical to the percentage for having a history of smoking tobacco cigarettes. Few
children had smoked cannabis by age 11 years (n=7; 1%). By age 17, the prevalence of
drug use was higher; an estimated 59% had a history of alcohol use; 45% had a history of
tobacco cigarette smoking; 35% had a history of cannabis use.

67

Table 4.2 Mean level of parental monitoring and percentages of drug use in
the sample. Data come from 774 children sampled from 1983-1985 newborn
discharge lists in southeast Michigan and contributing data to the analyses.
Age 11
Age 17
Parental Monitoring (Mean (SD))
Peer Alcohol Use (%)
Peer Tobacco Cigarette Smoking (%)
Peer Cannabis Smoking (%)
Ever Alcohol Use (%)
Ever Tobacco Cigarette Smoking (%)
Ever Cannabis Smoking (%)

36.1 (4.0)
19.6
28.0
19.6
10.1
10.6
1.0

68

--------58.9
44.7
34.7

4.3.3 Structural Equation Model for Parental Monitoring, Drug Use, and Peer Drug Use
at Age 11 and Drug Use at Age 17

Figure 4.1 presents the SEM with standardized path estimates for the 774 cohort
members, adjusted only for the stratification variables. All model fit indices were in the
good to excellent range (CFI = 0.976; TLI = 0.968; RMSEA = 0.037). As shown in
Figure 1, level of parental monitoring (PM 11), level of drug use (Drug 11), and level of
affiliation with drug using peers (DUP 11) were all correlated at age 11 (p<0.05). PM 11
was inversely related to both. Higher levels of PM 11 were associated with lower levels
of Drug 11 and DUP 11. Drug 11 and DUP 11 were correlated (p<0.05). Namely, higher
levels of Drug 11 were associated with higher levels of DUP 11. PM 11 predicted levels
of drug use at age 17 (Drug 17) (p=0.01). The negative estimate indicates that higher
levels of PM 11 were associated with lower levels of Drug 17. In addition, higher levels
of Drug 11 were predictive of higher levels of drug use at age 17 years (p<0.04). In
contrast, the association between level of DUP 11 and level of Drug 17 was not
statistically significant by conventional standards (p<0.30).

69

Figure 4.1 Unadjusted SEM predicting levels of drug use at age 17 (Drug 17) from levels of drug use (Drug 11), parental monitoring
(PM 11) and affiliation with drug using peers at age 11 (Drug11). Data come from 774 children sampled from 1983-1985 newborn
discharge lists in southeast Michigan and contributing data to the analyses. (*p<0.05)

70

Figure 4.2 displays results of the SEM after adjustment for sex and maternal
education. Model fit indices remained in the good to excellent range (= 0.948; TLI =
0.935; RMSEA = 0.040). Notably, only level of PM 11 was robustly associated with
level of Drug 17 in the adjusted model (p=0.01). The association between level of Drug
11 and level of Drug 17 was no longer statistically significant after adjustment for sex
(p=0.31).

71

Figure 4.2 Adjusted SEM predicting levels of drug use at age 17 (Drug 17) from levels of drug use (Drug 11), parental monitoring
(PM 11) and affiliation with drug using peers at age 11 (Drug11). Data come from 774 children sampled from 1983-1985 newborn
discharge lists in southeast Michigan and contributing to the analyses. (*p<0.05)

72

4.4 Discussion

Using a well-characterized 1983-1985 birth cohort, an estimate was derived for a
suspected causal influence of level of parental monitoring assessed at age 11 on level of
drug use at age 17, simultaneously probing other potential paths through levels of drug
use and affiliation with drug using peers at age 11 years. The findings from this study
can be summarized succinctly. First, at age 11 years, parental monitoring, level of
affiliation with drug using peers, and level of drug use were all robustly correlated with
one another. Second, parental monitoring at age 11 robustly predicted level of drug use
at age 17 years. Namely, higher levels of monitoring predicted lower levels of drug use.
Third, level of affiliation with drug using peers at age 11 did not predict level of drug use
at age 17. These findings lend considerable support to the “persistence effects”
hypothesis of parental monitoring on drug use; parental monitoring was related to drug
use at childhood and at the end of adolescence.
Before a more detailed discussion of the findings, several limitations should be
considered. Because parental monitoring and the questions used to construct the latent
variables (level of drug use at age 11 and level of affiliation with drug using peers) were
all assessed at age 11, the nature and direction of relationships between these correlated
variables is uncertain. For example, it might be the case, as some have suggested (Lloyd
& Anthony, 2003; Patterson et al., 1989), that parental monitoring reduces contact with
drug using peers, in turn lowering a child’s chance to use drugs. Unfortunately, these
data cannot disentangle the sequential processes of this relationship. In addition, because
parental monitoring was assessed at a single time point (age 11 years), it is unknown as to

73

whether it was stable or changed throughout the follow-up interval. Nonetheless,
parental monitoring at this one time point robustly predicted levels of drug use at age 17
years. Another limitation is that data on affiliation with drug using peers at ages other
than 11 years are unavailable in this sample. To the extent that affiliation with friends
who used drugs might have changed as the children matured, and whether and how these
changes in friends might have influenced later drug use, cannot be assessed in this study.
An additional limitation on causal inference is due to the observational nature of the data.
There is a large body of evidence documenting the role of parents and peers on
child and adolescent drug use. With respect to parenting practices, parental monitoring is
one key element that has been shown to be associated with child drug use. Specifically,
findings from previous research have consistently documented that higher monitoring is
associated with a lower risk of alcohol, tobacco, and other drug use in childhood and
adolescence (Chilcoat & Anthony, 1996; Chilcoat et al., 1995; DiClemente et al., 2001).
Using an SEM to simultaneously examine multiple influences of drug use in later
adolescence, the finding that higher levels of parental monitoring at age 11 predicted
lower levels of drug use at age 17 support and strengthen findings from prior research.
Moreover, this finding complements other similar findings from investigations examining
this association with common regression techniques, such as logistic regression. It also
provides evidence consistent with a durable influence of parental monitoring through the
ages when alcohol, tobacco, and cannabis use are most likely to occur. Collectively, the
findings lend support for the persistence hypothesis. This potential long-term influence
of parental monitoring could be due to two distinct possibilities which are not necessarily
mutually-exclusive: 1) children that are highly monitored at earlier ages might internalize

74

or carry with them the benefits of adept monitoring throughout development; 2) parental
monitoring is a persistent parental characteristic.
Previous studies have also reliably found that children who have drug using
friends or affiliate with deviant or drug using peers are more likely to use alcohol,
tobacco, and other drugs (Bauman & Ennett, 1996; Guo et al., 2002; van den Bree &
Pickworth, 2005). With respect to adolescent drug use, some have suggested that peers
might be more influential and more stable of an influence during adolescence (Guo et al.,
2002; van den Bree & Pickworth, 2005). In contrast, there are findings that resistance to
deviant peers is elevated in older adolescents. For example, Steinberg and Monahan
(2007) reported that resistance to peers increased in a linear fashion between ages 14 and
18 (Steinberg & Monahan, 2007). The findings from the present investigation support
this latter observation; i.e., the association between level of affiliation with drug using
peers and level of drug use in childhood, and no association between level of affiliation
with drug using peers in childhood and level of drug use in adolescence. There is also the
possibility that the influence of the age 11 peers “faded” over time. This “fade” might be
related to changes in peer group composition (i.e., peers at age 11 might not be the same
peers at age 17). Similarly, peers at age 16 or 17 might be more influential with respect
to drug use at age 17. Unfortunately, data on drug using peers at age 17 were
unavailable.
Findings from the present study have potential implications for future research.
Assessing parental monitoring, peer, and drug use pathways with more fine-grained
longitudinal data (e.g., even more frequent than year-by-year follow-up assessments)
would allow for the testing of hypothesized pathways from monitoring to affiliation with

75

drug using peers to later drug use. It would also make it possible for an examination of
stability and change in parental monitoring and its potential influence on drug use. In
addition, with respect to prevention efforts, booster sessions might not be needed if adept
parental monitoring can be established in the childhood years. Nonetheless, this idea is
speculative. The observed prediction will need to be replicated and tested in trials of
preventions aimed at improving parental monitoring.

76

5. CHAPTER FIVE - MANUSCRIPT THREE

LONGITUDINAL PATHS FROM PARENTAL MONITORING TO CANNABIS USE
IN CHILDHOOD AND ADOLESCENCE

ABSTRACT

Aims: To test the prospective association between parental monitoring (PM) and
subsequent recently active cannabis smoking, and to examine the hypothesized
meditational influence of deviant peer affiliation (DPA).
Design: A prospective longitudinal study completed within the context of a grouprandomized trial in a sample of grade school students.
Setting: A mid-Atlantic United States urban public school system.
Participants: 2,311 children, with assessments yearly from 1989-1994.
Measurements: Logistic regression models with generalized estimating equations (GEE)
were used to regress recently active cannabis smoking on the prior year’s assessment of
PM level, adjusting for other covariates. Structural equation modeling (SEM) of the
paths was used to examine the potential mediating influence of DPA.
Findings: Higher levels of PM predicted lower odds of being a recently active cannabis
smoker even after statistical adjustment for other covariates (adjusted odds ratio, AOR =
0.96; 95% confidence interval, CI = 0.92, 0.99). No appreciable mediation by level of
DPA was detected for the paths linking previous level of PM with later occurrence of
cannabis smoking.

77

Conclusions: The results from the present investigation help to shed new light on the
potential protective influence of parenting practices on later drug use. The findings
suggest that PM might help to prevent cannabis use throughout late childhood and early
adolescence. In addition, the predictive association between PM and cannabis use was
not appreciably mediated by level of DPA, in contradiction of a commonly held yet
rarely tested hypothesis.

78

5.1 Introduction

Cannabis use in childhood and adolescence is a public health concern. Although
alternative explanations have not been completely ruled out, mounting evidence links
early-onset cannabis use with increased risk for later adverse health and social outcomes,
including drug problems such as drug dependence and other psychiatric illnesses, as well
as educational achievement (Chen et al., 2005; Fergusson & Boden, 2008a; M. Lynskey
& Hall, 2000; Tien & Anthony, 1990; Wagner & Anthony, 2002a, 2002b). Therefore, in
order to identify potential preventive targets and inform intervention development, it is
important to gain a better understanding of the hypothesized paths that might lead to
more early-onset cannabis use.
Patterson et al. (1989) have proposed one comprehensive model, a developmental
progression for antisocial behavior, sometimes referred to as the social context model, to
explain a mechanism leading to child and adolescent antisocial behavior, including early
drug use (Patterson et al., 1989). The model, slightly refined by others based on
empirical findings, posits a mechanism in which poor parental monitoring (PM), defined
chiefly by tracking and knowledge of child whereabouts, leads to affiliation with deviant
peers, in turn leading to subsequent antisocial behavior such as drug use. Several
longitudinal studies have examined components of this model with respect to drug use.
For example, Lloyd and Anthony (2003) found that higher levels of PM were associated
with lower levels of affiliation with deviant peers across late childhood and early
adolescence (Lloyd & Anthony, 2003). Other longitudinal studies have observed
associations linking higher parental monitoring with reduced drug initiation (Chilcoat &
Anthony, 1996; Chilcoat et al., 1995). No epidemiologic study to date, however, has
79

tested the main components of the hypothesized model simultaneously within a
longitudinal context; i.e., simultaneously testing the hypothesized paths from PM to
affiliation with deviant peers to cannabis use.
Other salient gaps in the literature remain in the hypothesized influence of PM on
drug use. Notably, few studies have examined the specific suspected predictive
association between PM and cannabis use. However, as previously stated, cannabis use
in childhood and adolescence might be especially deleterious. Prior research has also
focused predominantly on drug initiation and has not considered whether PM is
associated with persistence of drug use or solely with initiation. Moreover, it is unknown
whether PM continues to exert an influence over time, regardless of whether a child has
initiated drug use in the past. In addition, previous investigations have been limited in
their examination of the link between PM and drug use to a relatively short time span,
usually no more than two or three years.
The present study uses a longitudinal study design and a well-characterized,
epidemiologically-credible sample of children from a school system in the mid-Atlantic
United States to help fill the existing gaps in the literature by testing the following aims.
First, the goal is to estimate the suspected influence of earlier levels of PM on later
cannabis use with data collected yearly over a six year study interval. Second, a test is
conducted to determine whether the association is consistent over the study interval or
whether it varies over time as children age. Third, direct paths from earlier levels of PM
to later cannabis smoking and indirect paths from earlier levels of monitoring to
subsequent levels of affiliation with deviant peers to later cannabis smoking are derived

80

and tested. That is, mediation by deviant peer affiliation (DPA) is tested in the
hypothesized predictive relation between PM and cannabis use.

5.2 Methods

5.2.1 Study Population and Sample

The present study builds on an epidemiology and prevention research program
initiated by Professors Sheppard Kellam, James C. Anthony, and their colleagues at the
Prevention Research Center of the Johns Hopkins University Bloomberg School of
Public Health. Detailed information on the program, research design and methods are
available elsewhere (Kellam & Anthony, 1998; Kellam et al., 1991) and are briefly
summarized here. The protocol for the research was reviewed and approved by the
review board for protection of human subjects in research at the Johns Hopkins
University Bloomberg School of Public Health. The data analysis protocol was also
reviewed and approved by the review board at Michigan State University.
In brief, data come from a prospective longitudinal study completed within the
context of a group-randomized trial of two interventions: a Good Behavior Game and a
Mastery of Learning curriculum (Kellam, Brown et al., 2008; Kellam, Reid, & Balster,
2008; Kellam et al., 1991). The study population was designated to include all first
graders entering 19 public elementary schools of a single urban school system located in
the mid-Atlantic United States during two successive school years (cohort 1 entering in
1985 and cohort 2 entering in 1986), with some schools designated as intervention

81

schools and matched schools designated as external controls. Sub-sampling did not take
place; efforts were made to recruit all incoming first-graders (n=2,311; cohort 1 = 1,196,
cohort 2 = 1,115). Specifically, there were external school controls as well as withinschool classroom controls. At the time of entry into first grade, children in each
intervention school were randomly assigned to either an intervention classroom or an
internal standard-setting classroom (control). Children entering external control schools
were assigned at random to these first grade classrooms.
The 19 participating schools were located within the catchment area of five preidentified urban areas of the school system. The composition of these areas encompassed
very poor to middle class families who were mainly non-Hispanic black and white.
Moreover, the 19 schools and city neighborhoods where the children were growing up
were conceptualized as an ecological niche. Therefore, follow-up assessments were
focused on the first-graders who remained in the same school system.
From 1985 through 1994, children were assessed via regular standardized face-toface interviews with teachers. Starting at third grade (Spring of 1989), and continuing
through Spring of 1994, there were standardized face-to-face interviews with the children
themselves, which covered a variety of health and behavioral outcomes, including drug
use. The present investigation focuses on the 1989-1994 child assessments, when the
children matured from middle-late childhood through early adolescence.
The number of students interviewed varied each year by the maximum number of
assessments allowed by the school during weeks in April-June, after school achievement
testing. In the Spring of 1994, the investigators decided to focus resources on the second
cohort of students who remained in the school system, and they only interviewed cohort

82

one students when cohort two students were not available. This explains the smaller
numbers of interviewed participants in 1994.

5.2.2 Main Measures

The following main interview measurements were taken each Spring from 1989
to 1994.

5.2.2.1 Parental Monitoring, PM

Level of PM was assessed via a child-reported 10-item scale. As described
elsewhere, the items were drawn from the Oregon Social Learning Center Parent
Monitoring Scale and adapted for age-appropriateness (Capaldi & Patterson, 1989;
Chilcoat et al., 1995; Lloyd & Anthony, 2003). The items in the scale encompass
parental supervision, tracking and knowledge of child whereabouts (e.g., “How often,
before you go out, do you tell your parents when you will be back?”). The complete scale
is included in Appendix D. Possible scores ranged from a low of 10 to a high of 41.

5.2.2.2 Recently Active Cannabis Smoking, CAN

Recently active (past-year) cannabis smoking (CAN) was assessed via a single
child-reported item: “When was the last time you smoked (marijuana)?”, with the local
term for marijuana inserted within the parentheses, as described elsewhere (Wilcox,

83

Storr, Benoit, & Anthony, 2005). Children were coded in relation to a dichotomous
variable; those who had smoked cannabis since Spring of the previous year (1) and those
who had not (0).

5.2.2.3 Deviant Peer Affiliation, DPA

Level of DPA was assessed via a five-item scale. As previously described, the
items were drawn from the Oregon Social Learning Center Peer Behavior Scale and
adapted for age-appropriateness (Capaldi & Patterson, 1989; Chilcoat et al., 1995; Lloyd
& Anthony, 2003). The five items assess friendships with children who participate in
deviant behaviors such as cheating on tests and hitting others. The scale is included in
Appendix E. Possible scores ranged from a low of five to a high of 25.

5.2.3 Other Covariates

Covariate values for sex, racial/ethnic minority status, and free/subsidized school
lunch status at first grade were drawn from a centralized school database at baseline, as
described elsewhere (Kellam & Anthony, 1998). Early aggression was measured upon
entry into primary school via the standardized teacher interview assessment, the Teacher
Observation of Classroom Adaptation-Revised (TOCA-R) (Lloyd & Anthony, 2003).
Tobacco smoking in the year prior to CAN was evaluated each year from 1989 to 1993
via the following child-reported assessment question: “When was the last time you
smoked tobacco?” For each year, a dichotomous variable was created to indicate those

84

who had smoked tobacco since the previous Spring (1) and those who had not (0).
Alcohol use in the year prior to CAN was evaluated each year from 1989 to 1993 in the
same manner; i.e., a dichotomous variable constructed from the assessment question:
“When was the last time you drank beer, wine or any other alcohol drink?” Cohort,
intervention group, and classroom/school were all design variables determined at
baseline. A dummy-coded indicator variable was created for each year of outcome
assessment.

5.2.4 Statistical Analysis

First, in order to estimate the time-series relationship linking prior level of PM
with the log-odds of CAN across the five outcome time points (i.e., 1990-1994), a logistic
regression approach, applying generalized estimating equations (GEE) was used (Zeger,
Liang, & Albert, 1988). Specifically, the log-odds of CAN was modeled as a function of
prior level of PM (i.e., the data were arrayed so that each row was time-lagged; the logodds of CAN for time point t was regressed on level of PM for time point t-1). Next, the
model was elaborated to include potentially confounding covariates. These covariates
included: sex, racial/ethnic minority status, free/subsidized school lunch status at first
grade, teacher-rated aggression in first grade, prior recent tobacco and alcohol use (i.e.,
time-lagged, t-1), cohort, intervention group, and year of outcome assessment. A few
children (n=11) tried cannabis before the time of initial PM assessment; a variable to
indicate this history of cannabis smoking was also introduced as a covariate. Productterms between the covariates and PM were also evaluated, but none qualified for entry at

85

the alpha level of 0.05. The GEE produces population-averaged estimates, taking into
account interdependencies of repeated observations for the same subject over time (Zeger
et al., 1988). For the present analysis, 1,448 individuals had available data for these GEE
analyses, which were performed using Stata 11 (StataCorp, 2009).
Second, alternating logistic regression (ALR) was used to check the estimates
obtained from the adjusted GEE model and examine the potential influence of clustering
at the classroom/school level (Bobashev & Anthony, 2000). ALR is similar to GEE in
that it takes into account interdependencies of correlated data and produces population
averaged estimates. Unlike GEE, however, ALR produces a directly interpretable
estimate of the amount of clustering in the data in the form of the pair-wise odds ratio
(PWOR). Furthermore, ALR is capable of handling two levels of clustering; here, the
individual and the classroom/school. The ALR model was fit using SAS 9.1 (SAS
Institute, Cary, NC).
Third, structural equation modeling (SEM)/path analysis was used for
simultaneous examination of prospective paths from level of PM to level of DPA and
onward to CAN. Following the steps outlined by Baron and Kenny (1986), three separate
models were estimated to examine mediation (Baron & Kenny, 1986). In the first model,
paths from level of PM at each time t to CAN at each time t+2 were estimated to test for
direct effects. The model also simultaneously adjusted for prior and subsequent
assessments of PM and CAN. In the second step, paths from level of PM at each time t
to level of DPA at each time t+1 were modeled to test for the predictive association
linking prior PM, the initial variable, with subsequent DPA, the hypothesized mediator.
In the second model, statistical adjustment was made for prior and subsequent

86

assessments of level of PM and level of DPA. In the comprehensive third model, effects
were estimated for the paths from level of PM at time t to level of DPA at time t+1 to
CAN at time t+2, as well as for the direct path from level of PM at t to CAN at t+2. The
third model also simultaneously adjusted for prior and subsequent assessments of PM,
DPA, and CAN. For all three models, SEM was implemented under maximum
likelihood estimation with robust standard errors (MLR) using Monte Carlo integration
with 1000 integration points in Mplus 6.0 (Muthén & Muthén, 1998-2010). The MLR
SEM approach in Mplus uses multiple imputation methods that allow for the inclusion of
subjects with incomplete data (Muthén & Muthén, 1998-2010). The SEM analyses were
conducted for 1,302 individuals with available data. Mplus produced linear regression
slopes for the meditational path when level of DPA was modeled as the outcome (i.e.,
Gaussian) and logistic regression slopes in the form of log-odds with corresponding odds
ratios when CAN was the outcome (i.e., binary).
Fourth, for an additional check of mediation, main predictive paths that were
statistically robust in all three models (i.e., where the paths were robust from PM to
recent cannabis use, PM to DPA, and PM to DPA to recent cannabis use) were tested for
mediation using the binary_mediation program with the bootstrap command with 500
replications in Stata 11 (StataCorp, 2009; UCLA). In accordance with Kenny’s approach
for testing mediation with dichotomous outcomes, the program standardizes the
coefficients before using the product of coefficients method to compute indirect effects
(Kenny, 2009; UCLA). This allows for the direct comparison of coefficients from one
model to another in computing the direct and indirect effects (Kenny, 2009; UCLA). It
should be noted, however, that some of the complexities of the data that are accounted for

87

in the SEM analyses (e.g., repeated measurements) are lost in the traditional regression
framework of the binary_mediation program.

5.3 Results

5.3.1 Characteristics of the Sample

Table 5.1 provides a description of key study sample characteristics at each year
of assessment. As shown in the table, at each time point there was a balance of
approximately 50% males and 50% females. A majority of the sample was from a
racial/ethnic minority subgroup (most of whom were African-American). As was
previously stated in the methods, most of the adolescents assessed in 1994 were from
cohort 2. The percentages of participants from each intervention group were similar
each year (~60% standard; ~20% Good Behavior Game; ~20% Mastery Learning).

88

Table 5.1 Characteristics of the sample at each assessment 1989-1994. Data come from 2,311 first graders enrolled in a
mid-Atlantic public school system and assessed yearly from 1989 through 1994.
Total Sample
1989
1990
1991
1992
1993
1994
n=2,311
n=1,530
n=1,233
n=1,543
n=1,416
n=1,251
n=816
Mean age range in years
8-9
9-10
10-11
11-12
12-13
13-14
n
%
n
%
n
%
n
%
n
%
n
%
n
%
Sex
Male
1,151 49.8 770 50.3 595 48.3 758 49.1 694 49.0 603 48.2 399 48.9
Female
1,160 50.2 760 49.7 638 51.7 785 50.9 722 51.0 648 51.8 417 51.1
Racial/ethnic
minority status
Nonminority
761
32.9 398 26.0 286 23.2 352 22.8 288 20.3 240 19.2 153 18.8
Minority
1,550 67.1 1,132 74.0 947 76.8 1,191 77.2 1,228 79.7 1,011 80.8 663 81.2
Cohort
1
1,196 51.8 784 51.2 639 51.8 762 49.4 705 49.8 632 50.5 269 33.0
2
1,115 48.2 746 48.8 594 48.2 781 50.6 711 50.2 619 49.5 547 67.0
Intervention group
Standard
1,339 57.9 883 57.7 688 55.8 892 57.8 816 57.6 712 56.9 477 58.5
Good behavior game
452
19.6 319 20.9 262 21.2 324 21.0 303 21.4 270 21.6 159 19.5
Mastery learning
520
22.5 328 21.4 283 23.0 327 21.2 297 21.0 269 21.5 180 22.0

89

5.3.2 Prevalence of CAN and Mean Scores for PM and DPA, 1989-1994

Table 5.2 depicts the prevalence (percentage) of CAN, the mean level of PM, and
the mean level of DPA for each year of assessment. CAN prevalence proportions
increased over the six years of assessment. Values were low for the first four years
(~1%) and higher for the final two years of assessment (4.3% and 13.4%, respectively).
The mean level of PM was relatively stable over the six assessment years, approximately
33 with a standard deviation of 5. The mean level of DPA was also relatively stable over
time, ranging from 9.9 to 11.1.

90

Table 5.2 Percentages of recent cannabis use and means scores of parental
monitoring and deviant peer affiliation for the assessments. Data come from a
cohort of 2,311 first graders enrolled in a mid-Atlantic public school system and
assessed yearly from 1989 through 1994.
Recent (past-year)
Deviant peer
cannabis use
Parental monitoring
affiliation
Assessment
year
n
%
SE
n
Mean SD
n
Mean SD
1989
1,506 0.7
0.2
1,435 32.2 5.2
1,513 11.1 3.9
1990
1,231 0.6
0.2
973
33.4 5.4
1,223
9.9
3.8
1991
1,541 1.2
0.3
1,361 33.2 5.0
1,536 10.4 3.9
1992
1,416 1.3
0.3
1,317 32.9 5.1
1,410 10.6 3.9
1993
1,247 4.3
0.6
1,163 32.8 5.1
1,251 10.9 3.9
1994
813
13.4 1.2
695
33.2 5.3
813
10.9 4.1
*The cannabis use value in 1989 is cumulative to the Spring of 1989; other estimates
are prevalence proportions reflecting use since the prior Spring.

91

5.3.3 Results from the GEE Logistic Regression on the Hypothesized PM-CAN Link

Tables 5.3 and 5.4 present results from the crude and adjusted GEE logistic
regression models, respectively. In the crude model, higher levels of PM predicted lower
odds of CAN (odds ratio, OR = 0.93; 95% confidence interval, CI = 0.90, 0.96).
Specifically, a one point increase on the PM scale signaled a seven percent decrease in
the odds of CAN. After adjustment for covariates, the estimated effect of PM on CAN
was slightly attenuated; however, it remained statistically robust (adjusted OR, AOR =
0.96; 95% CI = 0.92, 0.99). Because no statistical interactions were detected between
PM and any of the other covariates, the null hypothesis that the common odds ratio for
PM was uniform across subgroups could not be rejected. Notably, the predictive
association between level of PM and CAN did not vary by year of assessment.
The covariate-adjusted GEE logistic regression yielded other findings of interest
(Table 5.4). Males, children with higher levels of aggression in first grade, and prior
tobacco smoking and alcohol use (i.e., one year prior to CAN) predicted increased odds
of CAN. Cohort 2 children had lower odds of CAN than their older counterparts in
cohort 1. Compared with the first year of assessment, there was an excess occurrence of
CAN in 1993 and 1994.

92

Table 5.3 Results from the unadjusted GEE logistic regression
linking prior parental monitoring with odds of recent cannabis use.
Data come from 1,448 first graders enrolled in a mid-Atlantic
public school system and assessed yearly from 1989
through 1994.
Odds
95% Confidence
Ratio
Interval
p value
Parental monitoring in the
prior year
0.93
(0.90, 0.96)
<0.001
Sex
Female
Ref
Male
2.52
(1.67, 3.80)
<0.001
Race/ethnic minority
status
Nonminority
Ref
Minority
1.29
(0.78, 2.16)
0.32
Eligible for subsidized
lunch in first grade
No
Ref
Yes
1.06
(0.71, 1.57)
0.785
First grade aggression
1.35
(1.16, 1.58)
<0.001
Cannabis use at 1989
assessment
No
Ref
Yes
5.26
(1.59, 17.47)
0.007
Cigarette smoking in the
prior year
No
Ref
Yes
9.65
(6.27, 14.84)
<0.001
Alcohol use in the prior
year
No
Ref
Yes
4.62
(3.31, 6.46)
<0.001
Cohort
1
Ref
2
0.74
(0.51, 1.07)
0.113
Year of assessment
1990
Ref
1991
1.50
(0.59, 3.76)
0.393
1992
1.83
(0.76, 4.41)
0.179
1993
6.14
(2.96, 12.72)
<0.001
1994
23.72
(11.27, 49.92)
<0.001
Note: The analysis also adjusted for intervention group.
Data are suppressed because another investigator is responsible
for reporting on this variable.
93

Table 5.4 Results from the adjusted GEE logistic regression linking prior
parental monitoring with odds of recent cannabis use. Data come from
1,448 first graders enrolled in a mid-Atlantic public school system and
assessed yearly from 1989 through 1994.
Adjusted
95%
Odds
Confidence
p
Ratio
Interval
value
Parental monitoring in the prior year
0.96
(0.92, 0.99)
0.026
Sex
Female
Male
2.26
(1.45, 3.50) <0.001
Race/ethnic minority status
Nonminority
Minority
1.41
(0.82, 2.42)
0.211
Eligible for subsidized lunch in first
grade
No
Yes
1.05
(0.68, 1.62)
0.832
First grade aggression
1.28
(1.06, 1.55)
0.011
Cannabis use at 1989 assessment
No
Yes
6.15
(1.99, 18.98)
0.002
Cigarette smoking in the prior year
No
Yes
4.52
(2.48, 8.26) <0.001
Alcohol use in the prior year
No
Yes
2.07
(1.37, 3.15)
0.001
Cohort
1
2
0.49
(0.33, 0.73)
0.001
Year of assessment
1990
1991
1.07
(0.36, 3.22)
0.9
1992
1.39
(0.52, 3.73)
0.507
1993
4.63
(1.97, 10.90) <0.001
1994
23.83
(10.04, 56.54) <0.001
Note: The analysis also adjusted for intervention group.
Data are suppressed because another investigator is responsible
for reporting on this variable.

94

5.3.4 Examining Clustering at the Individual and Classroom/school Levels Via the ALR

Because children were sampled with respect to classroom/school, the extent to
which clustering would have an impact on the findings from the GEE logistic regression
was evaluated using ALR. Findings from the adjusted ALR are shown in Table 5.5. The
findings from ALR were nearly identical to the findings from the adjusted GEE logistic
regression. The estimated influence of parental monitoring on recent cannabis use was
the same (OR = 0.96; 95% CI = 0.92, 0.99). The pair-wise odds ratios (PWOR) depict
the amount of clustering on the individual and classroom/school levels. As shown in the
table, there was clustering on the individual level (PWOR = 3.18; 95% CI = 1.42, 7.11).
Clustering at the classroom/school level was attenuated after individual-level clustering
and other covariates were taken into account (PWOR = 0.90; 95% CI = 0.77, 1.06).

95

Table 5.5 Results from the ALR linking prior parental monitoring with odds of
recent cannabis use and accounting for clustering. Data come from 1,448 participants
originally recruited in 1985-1986 and followed-up from 1989-1994.
Adjusted Odds 95% Confidence
Ratio
Interval
p value
Parental Monitoring
0.96
(0.92, 0.99)
0.015
Sex
Female
Ref
Male
2.30
(1.55, 3.42)
<0.001
Race/ethnic minority status
Nonminority
Ref
Minority
1.34
(0.79, 2.29)
0.278
Eligible for subsidized lunch in
first grade
No
Ref
Yes
1.07
(0.69, 1.68)
0.759
First grade aggression
1.27
(1.04, 1.56)
0.02
Cannabis use at 1989 assessment
No
Ref
Yes
6.02
(1.93, 18.73)
0.002
Cigarette smoking in prior year
No
Ref
Yes
4.22
(2.36, 7.63)
<0.001
Alcohol use in prior year
No
Ref
Yes
2.04
(1.40, 2.97)
<0.001
Cohort
1
Ref
2
0.49
(0.33, 0.71)
<0.001
Year of assessment
1990
Ref
1991
1.15
(0.40, 3.27)
0.796
1992
1.45
(0.53, 3.95)
0.467
1993
4.86
(1.96, 12.05)
0.001
1994
26.31
(1.55, 11.09)
<0.001
Pair-wise
Clustering
Odds Ratio
95% CI
Individual
3.18
(1.42, 7.11)
CSS
0.90
(0.77, 1.06)
Note: Intervention group was also adjusted for but data are suppressed
because another investigator is responsible for reporting on this variable.

96

p value
0.005
0.227

5.3.5 Testing Hypothesized Mediation through DPA

Figure 5.1 depicts the SEM linking prior level of PM with subsequent CAN (two
years later). Specifically, the figure depicts the estimated lagged influence of level of PM
on CAN (i.e., the probability of CAN at each time t+2 regressed on level of PM at time t)
taking into account the other paths in the model. Paths from level of PM to CAN
indicated that higher monitoring was associated with lower odds of CAN at each
assessment; however, only the final two paths with CAN in 1993 (CAN93) and 1994
(CAN94) as the outcomes were robust (for CAN93: coefficient = -0.12, OR = 0.88,
p<0.05; for CAN94: coefficient = -0.10, OR = 0.90, p<0.05).

97

Figure 5.1 SEM estimates linking prior level of parental monitoring (PM) with recent cannabis use (CAN). Data come from 1,302 first
graders enrolled in a mid-Atlantic public school system and assessed yearly from 1989 through 1994.

*p<0.05

98

Figure 5.2 depicts the SEM evaluation of hypothesized paths from prior level of
PM to subsequent level of DPA. The figure shows a robust predictive association for
each path linking prior level of PM with subsequent level of DPA (coefficients ranging
from -0.05 to -0.08; all p<0.05). For each path, higher levels of PM were associated with
lower levels of DPA.

99

Figure 5.2 SEM estimates linking prior level of parental monitoring (PM) with level of deviant peer affiliation (DPA). Data come
from 1,302 first graders enrolled in a mid-Atlantic public school system and assessed yearly from 1989 through 1994.

*p<0.05

100

Figure 5.3 depicts the model extended to examine potential mediation via level of
DPA (i.e., paths from level of PM at t to level of DPA at t+1 to CAN at t+2). For each
sequence of the main paths of interest, level of PM was associated with level of DPA
(p<0.05), which was associated with subsequent CAN (p<0.05). Direct predictive paths
from level of PM to CAN remained inverse and robust for the final two paths (i.e., the
paths with CAN93 and CAN94 as the outcomes). Taken together with the findings from
the two previous figures, the results from Figure 3 provide evidence for slight mediation
by level of DPA in the association between level of PM and later CAN for the two final
main regression paths of interest (i.e., with CAN93 and CAN94 as the outcomes).
Nonetheless, the mediation is not appreciable. This is evidenced by the lack of sizeable
attenuation in the main direct predictive regression paths, from PM in 1991 (PM91) to
CAN93 and from PM in 1992 (PM92) to CAN94, between Figure 1 and Figure 3. For
example, the coefficient and corresponding OR for the path from PM92 to CAN94 was 0.10 and 0.90, respectively, in Figure 1, which was not appreciably distorted with the
addition of the suspected mediator, DPA in 1993, in Figure 3 (coefficient = -0.09, OR =
0.92).

101

Figure 5.3 SEM estimates testing paths from level of parental monitoring (PM) to level of deviant peer affiliation (DPA) to recent
cannabis use (CAN). Data come from 1,302 first graders enrolled in a mid-Atlantic public school system and assessed yearly from
1989 through 1994.

*p<0.05
102

Additional tests for mediation were then estimated for the two main paths that met
criteria for mediation in the SEMs (i.e., the path from PM91 to DPA92 to CAN93 and the
path from PM92 to DPA93 to CAN94. The total estimated indirect and direct effects are
shown in Table 5.6 for each main path. As shown in the table, none of the confidence
intervals entrap the null; therefore, all of the estimated effects are statistically robust. The
table also shows the estimated proportion of total effect mediated for each path. Both are
minimal, 0.31 for the path with CAN93 as the outcome and 0.17 for the path with
CAN94 as the outcome, indicating no appreciable mediation via level of DPA. That is,
level of DPA explained a very small proportion of the predictive association between
level of PM and recent cannabis use.

103

Table 5.6 Formal tests of mediation for the two main paths of interest that were robust in
the SEM. Data come from a cohort of 2,311 first graders enrolled in a mid-Atlantic
public school system and assessed yearly from 1989 through 1994.
Estimated path from PM91 to DPA92 to CAN93
Bias Corrected 95% Confidence
Estimate
Interval
Indirect effect
-0.07
(-0.11,-0.05)
Total indirect effect
-0.07
(-0.11, -0.05)
Direct effect
-0.16
(-0.29, -0.002)
Total effect
-0.23
(-0.35, -0.08)
Estimated proportion of total effect mediated = 0.31
Estimated path from PM92 to DPA93 to CAN94
Indirect effect
Total indirect effect
Direct effect
Total effect

Estimate
-0.05
-0.05
-0.23
-0.27

Estimated proportion of total effect mediated = 0.17

104

Bias Corrected 95% Confidence
Interval
(-0.08, -0.02)
(-0.08, -0.02)
(-0.34, -0.11)
(-0.39, -0.16)

5.4 Discussion

The findings from the present prospective study strengthen the evidence base for
the potential preventive influence of adept PM on later drug involvement in important
and novel ways. First, a robust predictive association linking higher levels of PM with
decreased odds of recent cannabis use was found. Specifically, every one point increase
on the monitoring scale signaled a four percent decrease in the odds of recent cannabis
use, even after statistical adjustment for other covariates. Second, variation by year of
assessment was not detected, indicating that the association between level of PM and
later cannabis use was relatively stable as the children aged. Third, in a test of the main
components of the model of developmental progression for antisocial behavior, or social
context theory, no appreciable mediation of the PM-CAN relationship by level DPA was
detected.
The findings from the present investigation are consistent with other longitudinal
studies that have observed a lower risk of drug use with increased levels of PM (Chilcoat
& Anthony, 1996; Chilcoat et al., 1995). Nonetheless, most of the prior evidence has
been limited to illegal drug initiation in general, and has not focused on cannabis
smoking in specific. The findings from the present investigation are consistent with the
few studies that have examined the relationship between PM and cannabis smoking in a
longitudinal context. For example, a prior prospective study conducted in Australia
found that poor PM was one of four predictors of adolescent drug use, including cannabis
(Hayatbakhsh et al., 2008). Similarly, in a population-based sample of adolescents in the
United States, the National Survey of Parents and Youth (NSPY), low PM was associated
with cannabis use (Martins et al., 2008). The results from the present study strengthen
105

and extend the findings from this prior work, offering the advantage of testing the
relationship between prior levels of PM and subsequent occurrence of CAN over multiple
yearly assessments. The results from the present investigation suggest that PM might
help to shield youths from smoking cannabis during the interval of late childhood to early
adolescence, and that the estimated influence of monitoring is stable over these years,
even as the incidence of CAN rises sharply during adolescence.
This is the first study of cannabis smoking that has examined the main processes
presented by Patterson et al. (1989) in the model of developmental progression for
antisocial behavior, sometimes called the social context theory. Prior evidence has
examined components of the theory (Patterson et al., 1989). Namely, Lloyd and Anthony
(2003) observed a robust association linking prior levels of PM with subsequent levels of
DPA (Lloyd & Anthony, 2003). Other studies have documented the association between
deviant peers and drug use (Kaplan, Martin, & Robbins, 1984) and PM and cannabis and
other drug use (Chilcoat & Anthony, 1996). No prior study has tested these paths
simultaneously with respect to cannabis smoking. At earlier ages, when CAN was less
common, the estimated direct effect for level of PM on the odds of CAN was not
statistically significant by conventional standards (i.e., at p<0.05). This finding might be
a result of low power to detect an association, as there were very few cannabis smokers at
earlier ages. Nonetheless, at later ages, when CAN became more common, level of DPA
did not appreciably mediate the predictive paths from level of PM to CAN. That is, the
direct paths from level of PM to subsequent CAN were not markedly attenuated when
level of DPA was included in the SEM. Therefore, the commonly held and argued belief
that the primary manner in which PM prevents drug use is by thwarting relationships with

106

deviant peers is not wholly supported by the evidence of this study. This finding is
foreshadowed in two previous studies on PM and tobacco initiation and PM and cannabis
initiation in a southeast Michigan cohort (Bohnert, Anthony, & Breslau; Bohnert et al.,
2009). In the two previous studies, having a tobacco using peer, a measure related to
DPA, was an important predictor of initiation; however, it did not appreciably attenuate
the PM-drug initiation link.
The possible implications of these observational findings should be considered in
light of the following potential limitations. First, causal interpretations are limited due to
the observational and self-reported nature of the data. Second, in the GEE analysis
statistical adjustment was made for a broad array of covariates; however, there remains
the possibility that some unmeasured variable related to both level of PM and cannabis
smoking might account for the observed association. Third, in the logistic GEE analysis
it is possible that the attempt to statistically control for a host of covariates might have led
to over-adjustment. This especially might be the case with respect to prior tobacco and
alcohol use (i.e., before cannabis smoking). These variables might be mediators on a
PM-influenced path to CAN and adjustment might not have been warranted. Fourth, due
to complexity and sample size constraints, we were not able to adjust for other covariates
in the SEM.
It is important to note that PM is only one component in a constellation of
parenting behaviors. Nonetheless, the cumulative evidence suggests that PM is a vital
component. The results from the present investigation help to shed new light on the
potential protective influence of parenting practices on cannabis smoking that starts
before mid-adolescence. The findings suggest that PM might help to prevent cannabis

107

smoking at every age in late childhood and early adolescence. In addition, contrary to a
conventionally held hypothesis, evidence from the present study suggests that the
predictive association linking PM with CAN is not appreciably mediated by level of
DPA. Replication of these findings in other settings and samples is necessary. The
potential influence of PM evaluated in relation to other facets of parenting is also needed.
Longitudinal studies with more finely-grained assessments might help to elucidate other
complex and varied mechanisms along the hypothesized path from PM to drug use and
beyond, perhaps testing the influence of PM on other later adverse health and social
outcomes that might be consequences of early cannabis smoking (Fergusson & Boden,
2008a, 2008b; D. B. Kandel, 1984; M. Lynskey & Hall, 2000; Tien & Anthony, 1990).
The length of the potential influence of PM remains unknown; i.e., when monitoring and
its potential influence effectively end. Population-based trials aimed at teaching parents
how to adeptly monitor their children are also needed (Spoth et al., 2009).

108

6. CHAPTER SIX

FINAL CONCLUSIONS

6.1 Summary of Findings

The study in Chapter Three estimated the predictive relationship linking parental
monitoring, as reported by children at age 11, with the cumulative incidence of cannabis
use up to age 17 years. Children with higher levels of parental monitoring were less
likely to initiate cannabis use from the ages of 11 to 17 years. Specifically, for every unit
increase on the parental monitoring scale there was a 4% reduction in the risk of cannabis
initiation, even after statistical adjustment for other covariates. In the study presented in
Chapter Three, children in the low monitoring group were consistently more likely than
children in the high monitoring group to initiate cannabis use at each age interval from 11
to 17 years.
The investigation in Chapter Four estimated the influence of level of parental
monitoring assessed at age 11 on level of drug use at age 17, simultaneously probing
potential paths through levels of drug use and affiliation with drug using peers at age 11.
The study found that level of parental monitoring, level of affiliation with drug using
peers, and level of drug use were all robustly correlated with one another at age 11 years.
Parental monitoring at age 11 predicted level of drug use at age 17; namely, higher levels
of monitoring predicted lower levels of drug use. In contrast, level of affiliation with
drug using peers at age 11 years did not robustly predict level of drug use at age 17 years.

109

In Chapter Five, a GEE logistic regression was used to estimate the association
between prior levels of parental monitoring and subsequent recently active cannabis
smoking. The findings indicated that for every one point increase on the monitoring
scale, there was a 4% decrease in the odds of recently active cannabis use, even after
statistical adjustment for other covariates. A statistical interaction between level of
parental monitoring and year of assessment was not detected, indicating that the
association between level of parental monitoring and odds of later cannabis use might not
vary across the study interval (i.e., from middle to late childhood through early
adolescence). In Chapter Five, potential mediation of the parental monitoring-cannabis
use path by level of affiliation with deviant peers was tested (i.e., the model of
developmental progression of antisocial behavior was tested). Although the results were
statistically significant, mediation by level of affiliation with deviant peers was minimal
and not appreciable.

6.2 Limitations

In both longitudinal samples used for the dissertation research, drug use was
assessed via standardized, self-reported items, which are subject to biases (e.g., recall and
reporting). Nevertheless, empirical evidence suggests that self-reported drug use data is
relatively accurate, especially in epidemiologic settings. For example, although Harrison
(1995) found evidence for a small underreporting bias in the National Household Survey
on Drug Abuse and the Monitoring the Future Survey, the author suggested that the
overall effects of the bias are small (Harrison, 1995). Nonetheless, it should be noted that

110

in other settings, such as the juvenile justice system, self-reported measured may be less
reliable (Magura & Kang, 1996).
Similarly, parental monitoring was collected via child self-repot, which could
have influenced the results. For example, children who use cannabis might report that
they are more highly monitored than those who do not use cannabis. The studies in the
present dissertation did try to overcome this limitation. For example, the possibility of
this circumstance occurring in the first manuscript was reduced via the exclusion of
children who had previously initiated cannabis use from the analysis. In the third study,
multiple measurements of parental monitoring and cannabis use were used, which might
have improved reliability. The best possible measurement for parental monitoring might
come from an independent observer (e.g., a fieldworker observing and recording
monitoring in the participants’ homes); however, the burden on participants, the length of
time a fieldworker would have to observe a given home, and the expense are prohibitive.
Nonetheless, future studies might benefit from using multiple informants of parental
monitoring.
In the present dissertation studies, there is the possibility that sample attrition
influenced the findings. Evidence for this prospect was not supported; there were no
differences on demographic characteristics between the full sample who participated in
baseline assessments and those from the sub-sample who participated in follow-up
assessments, for either of the two cohort samples used.
The data used for the studies on parental monitoring and cannabis use were
observational in nature; therefore, causal interpretations are cautioned. Randomized trials
aimed at increasing parental monitoring are needed. Nonetheless, with respect to

111

guidelines for epidemiologic causal criteria, the results do show a modest association,
temporality, a dose-response relationship, the findings have been replicated, and there is
plausibility and consistency.

6.3 Future Directions

The findings from the present dissertation research have potential implications for
future research. First, the duration of the potential influence of adept parental monitoring
on drug use is unknown. In Chapter Three and Chapter Four, higher parental monitoring
in childhood predicted lower occurrence of drug initiation and use into late adolescence.
Dishion and McMahon (1998) have hypothesized that the influence of parental
monitoring might extend to even later developmental intervals and older ages, namely,
young adulthood (Dishion & McMahon, 1998). However, empirical tests of that
hypothesis have not been conducted. Second, there has been less investigation on the
potential impact of parental monitoring on the possible sequelae of early cannabis and
other drug use. For example, it might be of interest to test if parental monitoring relates
to academic outcomes through early drug use. Third, the finding in Chapter Five of no
appreciable mediation by level of deviant peer affiliation needs to be replicated in other
studies. If the findings are replicated in other samples, other potential mechanisms in the
parental monitoring-drug use association will need to be explored. Fourth, new evidence
is emerging with respect to the complex interrelationships between parental monitoring,
genetic predisposition, and other environmental influences on externalizing and smoking
behaviors (Dick et al., 2009; Dick et al., 2007). For example, using a twin sample, Dick

112

and colleagues (2007) observed that environmental influences were more important for
adolescent smoking when children were highly monitored but genetic influences
accounted for greater explanatory power at low parental monitoring levels (Dick et al.,
2007). These findings will need to be replicated in other diverse settings. It is unknown
if this observation might extend to other drugs or drug use disorders. Moreover, as new
candidate genes for drug use and dependence emerge, it will be necessary to test for
potential interactions with environmental influences, such as parental monitoring. Fifth,
longitudinal studies with more finely-grained assessments might help to elucidate other
complex and varied mechanisms along the paths from parental monitoring to drug use,
and could help explain the mechanisms underlying the association between parental
monitoring and drug use. Sixth, with the exception of a few recent noteworthy studies
(e.g. (Spoth et al., 2009)), there is a dearth of epidemiologic intervention trials aimed at
improving parenting with the goal of preventing early drug use. Most of the work to this
point has been in schools; however, new trials might aim at improving parental
monitoring in other study settings, e.g., community settings. These trials should also test
outcomes on suspected paths to early drug use, such as affiliation with deviant peers, and
possibly from early drug use to other potential consequences of early use. Seventh, most
investigations of the potential protective influence of parental monitoring have been
conducted in the United States. It is unknown if the generally observed associations in
the US-based research would hold in other countries.

113

6.4 Conclusions

Parental monitoring is one component in a constellation of parenting behaviors.
Nevertheless, the cumulative evidence suggests that parental monitoring is a vital
parenting element, and might shield youths from a host of delinquent behavior, including
cannabis and other drug use. Consistent with evidence from prior studies, the findings
from the present research help confirm the theory that parenting and familial
characteristics might exert long-lasting influences on a child’s risk of initiating and using
illegal drugs. The influence of parental monitoring was durable over the interval from
childhood to the end of adolescence, covering the ages when cannabis initiation and use
is most likely to occur. No appreciable mediation of the parental monitoring-cannabis
use association was detected by level of deviant peer affiliation; i.e., the direct effect
from parental monitoring to cannabis use remained relatively unchanged in the
meditational models that included a path through deviant peer affiliation. If future
prevention trials aimed at improving parental monitoring show efficacy, the interventions
used in the trials might be relatively inexpensive, mass-action methods for reducing drug
initiation and use before adulthood.

114

APPENDICIES

115

APPENDIX A

Technical Appendix for Chapter Three

In this study the research design is a prospective cohort with three waves of
assessment. The study population was designated to include low birth weight (LBW) and
normal birth weight (NBW) newborns from southeast Michigan. The sampling approach
involved randomly selecting LBW and NBW children from 1983-1985 newborn
discharge lists of two hospitals in southeast Michigan, one serving an inner-city
community and the other serving a suburban community. Forty-seven children with
severe neurologic impairments were excluded from the initial sample. The resulting
sample consisted of 1095 children. Of the 1095 children eligible for the study, 823
(75.2%) participated in the initial assessment from 1990 to 1992, when they were six
years of age. Follow-up assessments were conducted at 11 years (n=717; 87.1%) and 17
years of age (n=713; 86.6%). Six hundred and fifty-seven children completed both age
11 and age 17 follow-up assessments. Some of the designated participants for this study
(i.e., the 657 children) had missing or invalid responses to key study variables. For this
reason, the effective sample size for the present investigation and the proportion of
designated participants with useable data are 641 and 77.9%, respectively. The study
protocol was reviewed and approved by the cognizant institutional review board for
protection of human subjects in research.
The key response variable for the study was the cumulative incidence of cannabis
use up to age 17 years. It was assessed during the age 17 assessment via youth self-report

116

using the following standardized dichotomous (coded Yes (1) or No (0)) question about
cannabis use: “Have you ever, even once, used marijuana or hashish?” Data from the age
11 assessment were used to identify children who had initiated cannabis smoking by age
11 and were therefore no longer at risk for initiation during the 11 to 17 years age-span.
In addition, for those respondents who had initiated cannabis by age 17, a follow-up
question queried the age of their first cannabis use.
The suspected causal determinant of interest was level of parental monitoring. It
was assessed by child self-report at age 11 via a standardized 10-item scale (the full scale
is attached in Appendix D) (Capaldi & Patterson, 1989; Chilcoat et al., 1995). The items
encompass child supervision and tracking of activities outside the school environment
(e.g., whether an adult was present within 1 hour of the child arriving home from school,
how often the child talked with the parents about plans for the coming day, and whether
the child knew how to contact the parents if they were not at home after school). On
seven of the items, children responded on a five-point scale ranging from All of the time
(1) to Never (5); responses were coded either Clear (1) or Unclear (2) on two items; and
on a single item the responses were coded Yes (1) or No (2). A parental monitoring score
was constructed by reversing the coding and summing the scores on the 10 items.
Possible scores ranged from a low of 10 to a high of 41.
The guiding conceptual model was one in which the risk of cannabis initiation
from age 11 to 17 years was predicted by level of parental monitoring at age 11 years.
The plan for data analysis was organized in relation to standard "explore, analyze,
explore" cycles, in which the first exploratory steps involve Tukey-style box-and-whisker
plots and other exploratory data analyses to shed light on the underlying distributions of

117

each response variable and covariate of interest. In the initial analysis step, the task was
to estimate the cumulative occurrence of cannabis initiation from ages 11 to 17 years.
First, using a method described by Zou (2004), unadjusted and adjusted Poisson
regressions with robust error variances were conducted to estimate the predictive
association linking parental monitoring at age 11 years with the initiation of cannabis use
from the ages of 11 to 17 years (Zou, 2004). Estimates using this method have been
shown to yield more precise estimates of relative risk than odds ratios derived in
traditional logistic regression, especially when the outcome of interest occurs in greater
than 10% of the sample (i.e., when estimated odds ratios may not approximate relative
risk) (Zou, 2004). The general equation for the Poisson regression is given by the
equation:

𝑙𝑜𝑔 [𝜋(𝑥 𝑖 )] = 𝛼 + 𝛽𝑥 𝑖 + …

where 𝜋(𝑥 𝑖 ) is the probability of the outcome; β is the slope of the main covariate of
interest. The exponentiation of β gives the relative risk, and the robust sandwich

estimator is used to give the appropriate variance. Male-female subgroup variation in the
parental monitoring-cannabis initiation association was evaluated via product terms, as
was subgroup variation associated with race. None were detected at the alpha level of
0.05.
For a subsidiary survival analysis, a dichotomous variable was constructed from
the original parental monitoring scale using the median as the cutoff (low parental
monitoring (<37) and high parental monitoring (≥37)). Next, Kaplan-Meier curves were
derived for the above- and below-median monitoring groups in order to inspect the
failure rates of first cannabis use, year by year from age 11 to 17 years. Specifically, the

118

approach involved specifying the elapsed time from the age at the age 11 assessment until
the age of first cannabis use. Adolescents who never used cannabis contributed personyears up to the time of their age at the age 17 interview. A log-rank test was conducted to
formally test whether the two survival curves differed from one another (alpha = 0.05).
In a post-estimation exploratory step, the method of plotting fractional
polynomials, as described by Royston and Altman (1994), was employed to probe into
the issue of possible non-linearity in the parental monitoring – cannabis initiation
association (Royston & Altman, 1994). All analyses were conducted using Stata 11
(StataCorp, 2009). In this work, precision of the study estimates are stressed with a focus
on 95% confidence intervals; p-values are presented as an aid to interpretation.

119

APPENDIX B

Technical Appendix for Chapter Four

In this study the research design is a prospective cohort with three waves of
assessment. The study population was designated to include low birth weight (LBW) and
normal birth weight (NBW) newborns from southeast Michigan. The sampling approach
involved randomly selecting LBW and NBW children from 1983-1985 newborn
discharge lists of two hospitals in southeast Michigan, one serving an inner-city
community and the other serving a suburban community. Forty-seven children with
severe neurologic impairments were excluded from the initial sample. The resulting
sample consisted of 1095 children. Of the 1095 children eligible for the study, 823
(75.2%) participated in the initial assessment from 1990 to 1992, when they were six
years of age. Follow-up assessments were conducted at 11 years (n=717; 87.1%) and 17
years of age (n=713; 86.6%). Of the original cohort of 823 children, 774 (94.0%) who
had information on at least one outcome variable contributed data to the present
investigation. The institutional review boards of the participating institutions approved
the study.
There were several key variables for this study. Parental monitoring (PM 11) was
assessed via child self-report at 11 years of age using a standardized 10-item scale
(Appendix D) (Capaldi & Patterson, 1989; Chilcoat et al., 1995). The items encompass
child supervision and tracking of activities outside the school environment (e.g., whether
an adult was present within 1 hour of the child arriving home from school, how often the
child talked with the parents about plans for the coming day, and whether the child knew
120

how to contact the parents if they were not at home after school). On seven of the items,
children responded on a five-point scale ranging from All of the time (1) to Never (5);
responses were coded either Clear (1) or Unclear (2) on two items; and on a single item
the responses were coded Yes (1) or No (2). A summary score was generated by
reversing the coding and summing the scores. Scores ranged from a low of 10 to a high
of 41, with higher scores indicating higher levels of monitoring.
Level of drug use at age 11 (Drug 11) was constructed as a continuous latent
variable from the following standardized, child-reported age of onset questions which
were coded into dichotomous used/never used variables for alcohol, tobacco cigarette
smoking and cannabis smoking, respectively: (1) “How old were you the first time you
drank beer, wine, wine coolers, liquor, or any other drink with alcohol in it?”; (2) “How
old were you when you first smoked a tobacco cigarette, even just a puff?”; (3) “How old
were you the first time you smoked (marijuana/reefer)?”
Level of affiliation with drug using peers at age 11 (DUP 11) was constructed as a
continuous latent variable from the following standardized, child-reported dichotomous
(yes/no) questions on peer alcohol, tobacco cigarette smoking, and cannabis smoking,
respectively: (1) “Do you have any friends around your age who ever drink alcohol?”; (2)
“Do you have any friends around your age who ever smoke tobacco cigarettes?”; (3) “Do
you have any friends around your age who ever smoke (marijuana/reefer)?”
Level of drug use at age 17 (Drug 17) was constructed as a continuous latent
variable from the following three standardized, child-reported dichotomous (yes/no)
questions on alcohol, tobacco cigarette smoking and cannabis smoking, respectively: (1)
“Have you ever, even once, had a drink of any type of alcoholic beverage? Do not

121

include sips from another person’s drink.”; (2) “Have you ever smoked a cigarette, even
one or two puffs?”; (3) “Have you ever, even once, used marijuana or hashish?”
Structural equation modeling (SEM) was used to simultaneously test paths of
PM11 with DUP11 and Drug 11 and 17. All models were fit in Mplus 6.0 using the
weighted least squares means and variance adjusted estimator (Muthén & Muthén, 19982010). The latent variable SEM approach provides several advantages over traditional
regression techniques. First, it allows for simultaneous testing of multiple regression
paths of complex relationships. Second, common sources of variation between outcome
variables are taken into account and the path estimates are unencumbered by assumptions
about error. Third, Mplus handles missing data using multiple imputation methods to use
all available data (Muthén & Muthén, 1998-2010). Model fit was assessed via the
Comparative Fit Index (CFI) (where good model fit is denoted by a score >0.90) (Bentler,
1990), the Tucker-Lewis Index (TLI) (where good model fit is denoted by a score
>0.90)(Bentler & Bonett, 1980), and the Root Mean Square Error of Approximation
(RMSEA) (where good model fit is denoted by a score <0.05) (Bentler, 1990).

122

APPENDIX C

Technical Appendix for Chapter Five

The present study builds on an epidemiology and prevention research program
initiated by Professors Sheppard Kellam, James C. Anthony, and their colleagues at the
Prevention Research Center of the Johns Hopkins University Bloomberg School of
Public Health. The data come from a prospective longitudinal study completed within
the context of a group-randomized trial. At the time of entry into first grade, children
were randomly assigned to either an intervention classroom, with a Good Behavior Game
program or Mastery Learning Program, or to a standard classroom. The study population
was designated to include all first graders entering 19 public elementary schools of a
single urban school system located in the mid-Atlantic United States during two
successive school years (cohort 1 entering in 1985 and cohort 2 entering in 1986). Subsampling did not take place; efforts were made to recruit all incoming first-graders
(n=2,311; cohort 1 = 1196, cohort 2 = 1115).
The 19 participating schools were located within the catchment area of five preidentified urban areas of the school system. The composition of these areas encompassed
very poor to middle class families who were mainly non-Hispanic black and white.
Moreover, the 19 schools and city neighborhoods where the children were growing up
were conceptualized as an ecological niche. Therefore, follow-up assessments were
focused on the first-graders who remained in the niche.

123

From 1985 through 1994, children were assessed via regular standardized face-toface interviews with teachers. Starting at third grade (1989), and continuing through
1994, there were standardized face-to-face interviews with the children themselves,
which covered a variety of health and behavioral outcomes, including drug use. The
present investigation focuses on the child assessments from 1989 through 1994, when the
children ranged in age from 8 and 9 to 13 and 14 years old.
Each year in the Spring from 1989 through 1993, the interview was repeated. The
number of students interviewed varied each year by the maximum number of assessments
allowed by the school that respective Spring. In the Spring of 1994, the investigators
decided to focus resources on the second cohort of students who remained in the school
system, and they only interviewed cohort 1 students when cohort 2 students were not
available. Therefore, there were lower numbers of participants in 1994. The sample
sizes for each year from 1989 through 1994 were: 1530, 1233, 1543, 1416, 1251, and
816, respectively.
The main outcome of interest was recently active (past-year) cannabis smoking.
It was assessed in the interviews every spring from 1990 through 1994 via a single childreported item: “When was the last time you smoked (marijuana/Reefer)?” Children were
coded into a dichotomous variable; those who had used cannabis since June 1 of the
previous year (1) and those who had not (0). Past-year use in 1990 was coded to include
individuals who were new initiates and did not use cannabis in 1989.
The main suspected causal determinant was level of parental monitoring. Parental
monitoring was assessed each Spring from 1989 through 1994 via a standardized, childreported 10-item scale. As described elsewhere, the items were drawn from the Oregon

124

Social Learning Center Parent Monitoring Scale and adapted for age-appropriateness
(Capaldi & Patterson, 1989; Chilcoat et al., 1995). The items in the scale encompass
parental supervision, tracking, and knowledge of child whereabouts (e.g. “How often,
before you go out, do you tell your parents when you will be back?”). The complete scale
is included in Appendix D. Possible scores ranged from a low of 10 to a high of 41.
Another central variable of interest was level of deviant peer affiliation. The
variable was assessed each Spring from 1989 through 1994 via a standardized, childreported five-item scale. As previously described, the items were drawn from the Oregon
Social Learning Center Peer Behavior Scale and adapted for age-appropriateness
(Capaldi & Patterson, 1989; Chilcoat et al., 1995; Lloyd & Anthony, 2003). The five
items assess friendships with children who participate in deviant behaviors such as
cheating on tests and hitting others. The scale is included in the appendix (Appendix B).
Possible score ranged from a low of 5 to a high of 25.
The present investigation employed several analytic steps. First, a logistic
regression approach was used, applying generalized estimating equations (GEE), to
estimate the time-series relationship linking prior level of parental monitoring with the
probability of recent cannabis use across the five outcome time points (i.e., 1990-1994)
(Zeger et al., 1988). The model is a population-averaged model. The general equation is
as follows:

𝑝

𝑙𝑜𝑔𝑖𝑡 �𝜇 𝑖𝑗 � = 𝛽0 + � 𝛽ℎ 𝑋ℎ
ℎ=1

where 𝜇 𝑖𝑗 is the marginal probability of the outcome; 𝛽0 is the intercept; 𝛽ℎ is the slope
for a given covariate of interest from the array of covariates under study. In the present

125

study, the log-odds of recently active cannabis use was modeled as a function of prior
parental monitoring (i.e. the data was set up so that each row was time-lagged; the logodds of recently active cannabis use for time point t was regressed on parental monitoring
for time point t-1). Next, the model was elaborated to include potentially confounding
covariates. These covariates included: sex, racial/ethnic minority status, free/subsidized
school lunch status at first grade, teacher-rated aggression in first grade, prior recent
tobacco and alcohol use (i.e., time-lagged, t-1), cohort, intervention group, and year of
outcome assessment. A few children (n=11) tried cannabis before the time of initial PM
assessment; a variable to indicate this history of cannabis smoking was also introduced as
a covariate. Product-terms between the covariates and parental monitoring were also
evaluated. None were detected at the alpha level of 0.05. The GEE produces populationaveraged estimates; taking into account interdependencies of repeated observations for
the same subject over the survey assessment time points. It also offers the advantage of
using all available data (Zeger et al., 1988). For the present analysis, 1,448 individuals
had available data for these GEE analyses, which were performed using Stata 11
(StataCorp, 2009).
Second, the method of alternating logistic regression (ALR) was used to check the
estimates obtained from the adjusted GEE model and examine the potential influence of
clustering at the classroom/school level (Bobashev & Anthony, 2000). ALR is similar to
GEE in that it takes into account interdependencies of correlated data and produces
population averaged estimates. Unlike GEE, however, ALR produces a directly
interpretable estimate of the amount of clustering in the data in the form of the pair-wise
odds ratio (PWOR). The PWOR is analogous to the cross-product ratio in a two-by-two

126

table. Furthermore, ALR is capable of handling two levels of clustering; here, the
individual and the classroom/school. The general equation for a two level nested design
is as follows:

log�𝑃𝑊𝑂𝑅 𝑗𝑘𝑙𝑚 � = 𝛼0𝑍0 𝑗𝑘 + 𝛼1𝑍1 𝑙𝑚

where Z0jk is 1 if the pair (j,k) belongs to the same cluster (classroom/school) or 0 if not,
and Z1lm is 1 if the pair (l,m) belongs to the same subcluster (individual) or 0 if not.
Exponentiation of (α0) and (α0 + α1) gives the PWORs for the cluster and subcluster,
respectively. The ALR model was fit using SAS 9.1 (SAS Institute, Cary, NC).
Third, structural equation modeling (SEM)/path analysis was used for
simultaneous examination of prospective paths from level of PM to level of DPA and
onward to CAN. Following the steps outlined by Baron and Kenny (1986), three separate
models were estimated to examine mediation (Baron & Kenny, 1986). In the first model,
paths from level of PM at each time t to CAN at each time t+2 were estimated to test for
direct effects. The model also simultaneously adjusted for prior and subsequent
assessments of PM and CAN. In the second step, paths from level of PM at each time t
to level of DPA at each time t+1 were modeled to test for the predictive association
linking prior PM, the initial variable, with subsequent DPA, the hypothesized mediator.
In the second model, statistical adjustment was made for prior and subsequent
assessments of level of PM and level of DPA. In the comprehensive third model, effects
were estimated for the paths from level of PM at time t to level of DPA at time t+1 to
CAN at time t+2, as well as for the direct path from level of PM at t to CAN at t+2. The
third model also simultaneously adjusted for prior and subsequent assessments of PM,
DPA, and CAN. For all three models, SEM was implemented under maximum

127

likelihood estimation with robust standard errors (MLR) using Monte Carlo integration
with 1000 integration points in Mplus 6.0 (Muthén & Muthén, 1998-2010). The MLR
SEM approach in Mplus uses multiple imputation methods that allow for the inclusion of
subjects with incomplete data (Muthén & Muthén, 1998-2010). The SEM analyses were
conducted for 1,302 individuals with available data. Mplus produced linear regression
slopes for the meditational path when level of DPA was modeled as the outcome (i.e.,
Gaussian) and logistic regression slopes in the form of log-odds with corresponding odds
ratios when CAN was the outcome (i.e., binary).
Fourth, main predictive paths that were statistically robust in all three models (i.e.,
where the paths were robust from PM to recent cannabis use, PM to DPA, and PM to
DPA to recent cannabis use) were tested for mediation using the binary_mediation
program with the bootstrap command with 500 replications in Stata 11 (StataCorp, 2009;
UCLA). In accordance with Kenny’s approach for testing mediation with dichotomous
outcomes, the program standardizes the coefficients before using the product of
coefficients method to compute indirect effects (Kenny, 2009; UCLA). This allows for
the direct comparison of coefficients from one model to another in computing the direct
and indirect effects (Kenny, 2009; UCLA). It should be noted, however, that some of the
complexities of the data that are accounted for in the SEM analyses (e.g., repeated
measurements) are lost in the traditional regression framework of the binary_mediation
program.

128

APPENDIX D

Parental Monitoring Scale
1. What time does your mom or dad or parents expect you to come home from
school? (clear, unclear)
2. What time do they expect you to come home on a weekend night? (clear, unclear)
3. How often would your mom or dad or parents or a sitter know if you came home
an hour late on weekends? (always, most of the time, sometimes, hardly ever,
never)
4. Are there kids your mom or dad or parents don't allow you to play with? (always,
most of the time, sometimes, hardly ever, never)
5. How often, before you go out, do you tell your parents when you will be back?
(always, most of the time, sometimes, hardly ever, never)
6. If your parents or a sitter are not at home, how often do you leave a note for them
about where you are going? (always, most of the time, sometimes, hardly ever,
never)
7. How often do you check in with your parents or a sitter after school before going
to play? (always, most of the time, sometimes, hardly ever, never)
8. When you get home from school, how often is someone there within 1 hour?
(always, most of the time, sometimes, hardly ever, never)
9. If you are at home when your parents are not, how often do you know how to get
in touch with them? (always, most of the time, sometimes, hardly ever, never)
10. How often do you talk with your mom or dad or parents about your plans for the
coming day? For instance, talk about what will happen at school or with friends?
(always, most of the time, sometimes, hardly ever, never)

129

APPENDIX E

Deviant Peer Affiliation Scale
1. During the last school year, how many of your friends have cheated on school
tests? (none, very few of them, some of them, most of them, all of them)
2. During the last year, how many of your friends have ruined or damaged
something on purpose that wasn’t theirs? (none, very few of them, some of them,
most of them, all of them)
3. During the last year, how many of your friends have stolen something worth less
than five dollars? (none, very few of them, some of them, most of them, all of
them)
4. During the last year, how many of your friends have hit or threatened to hit
someone without any real reason? (none, very few of them, some of them, most of
them, all of them)
5. During the last year, how many of your friends have suggested that you do
something against the law? (none, very few of them, some of them, most of them,
all of them)

130

REFERENCES

131

REFERENCES

Andreasson, S., Allebeck, P., Engstrom, A., & Rydberg, U. (1987). Cannabis and
schizophrenia. A longitudinal study of Swedish conscripts. Lancet, 2(8574),
1483-1486.
Anthony, J. C. (2002). Death of the 'stepping-stone' hypothesis and the 'gateway' model?
Comments on Morral et al. Addiction, 97(12), 1505-1507.
Anthony, J. C. (2006). The epidemiology of cannabis dependence. In R. A. Roffman &
R. S. Stephens (Eds.), Cannabis dependence: its nature consequences and
treatment. New York: Cambridge University Press.
Anthony, J. C., & Van Etten, M. L. (1998). Epidemiology and its rubrics. In A. S.
Bellack & M. Hersen (Eds.), Comprehensive clinical psychology (First ed.). New
York: Pergamon.
Anthony, J. C., Warner, L. A., & Kessler, R. C. (1994). Comparative epidemiology of
dependence on tobacco, alcohol, controlled substances, and inhalants: Basic
findings from the national comorbidity survey. Experimental and Clinical
Psychopharmacology, 2(3), 244-268.
Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in
social psychological research: conceptual, strategic, and statistical considerations.
J Pers Soc Psychol, 51(6), 1173-1182.
Bauman, K. E., & Ennett, S. T. (1996). On the importance of peer influence for
adolescent drug use: commonly neglected considerations. Addiction, 91(2), 185198.
Bentler, P. M. (1990). Comparative fit indexes in structural models. Psychol Bull, 107(2),
238-246.
Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness of fit in the
analysis of covariance structures. Psychological Bulletin, 88(3), 588-606.
Block, L. G., Morwitz, V. G., Putsis, W. P., Jr., & Sen, S. K. (2002). Assessing the
impact of antidrug advertising on adolescent drug consumption: results from a
behavioral economic model. Am J Public Health, 92(8), 1346-1351.
Bobashev, G. V., & Anthony, J. C. (1998). Clusters of marijuana use in the United States.
Am J Epidemiol, 148(12), 1168-1174.
Bobashev, G. V., & Anthony, J. C. (2000). Use of alternating logistic regression in
studies of drug-use clustering. Subst Use Misuse, 35(6-8), 1051-1073.
132

Bohnert, K. M., Anthony, J. C., & Breslau, N. Parental monitoring at age 11 and
subsequent onset of cannabis use up to age 17: results from a prospective study.
Under Review.
Bohnert, K. M., Rios-Bedoya, C. F., & Breslau, N. (2009). Parental monitoring at age 11
and smoking initiation up to age 17 among Blacks and Whites: a prospective
investigation. Nicotine Tob Res, 11(12), 1474-1478.
Botvin, G. J., Baker, E., Dusenbury, L., Botvin, E. M., & Diaz, T. (1995). Long-term
follow-up results of a randomized drug abuse prevention trial in a white middleclass population. JAMA, 273(14), 1106-1112.
Botvin, G. J., Griffin, K. W., Diaz, T., Scheier, L. M., Williams, C., & Epstein, J. A.
(2000). Preventing illicit drug use in adolescents: long-term follow-up data from a
randomized control trial of a school population. Addict Behav, 25(5), 769-774.
Breslau, N., Brown, G. G., DelDotto, J. E., Kumar, S., Ezhuthachan, S., Andreski, P., et
al. (1996). Psychiatric sequelae of low birth weight at 6 years of age. J Abnorm
Child Psychol, 24(3), 385-400.
Brook, J. S., Adams, R. E., Balka, E. B., & Johnson, E. (2002). Early adolescent
marijuana use: risks for the transition to young adulthood. Psychol Med, 32(1),
79-91.
Brook, J. S., Richter, L., & Whiteman, M. (2000). Effects of parent personality,
upbringing, and marijuana use on the parent-child attachment relationship. J Am
Acad Child Adolesc Psychiatry, 39(2), 240-248.
Brook, J. S., Whiteman, M., Cohen, P., & Tanaka, J. S. (1992). Childhood precursors of
adolescent drug use: a longitudinal analysis. Genet Soc Gen Psychol Monogr,
118(2), 195-213.
Brook, J. S., Whiteman, M., Gordon, A. S., & Brenden, C. (1983). Older brother's
influence on younger sibling's drug use. J Psychol, 114(1st Half), 83-90.
Brook, J. S., Whiteman, M., Gordon, A. S., & Brook, D. W. (1985). Father's influence on
his daughter's marijuana use viewed in a mother and peer context. Adv Alcohol
Subst Abuse, 4(3-4), 165-190.
Capaldi, D., & Patterson, G. R. (1989). Psychometric properties of fourteen latent
constructs from the Oregon Youth Study. New York: Springer-Verlag.
Chen, C. Y., O'Brien, M. S., & Anthony, J. C. (2005). Who becomes cannabis dependent
soon after onset of use? Epidemiological evidence from the United States: 20002001. Drug Alcohol Depend, 79(1), 11-22.
133

Chen, C. Y., Wagner, F. A., & Anthony, J. C. (2002). Marijuana use and the risk of
Major Depressive Episode. Epidemiological evidence from the United States
National Comorbidity Survey. Soc Psychiatry Psychiatr Epidemiol, 37(5), 199206.
Chilcoat, H. D., & Anthony, J. C. (1996). Impact of parent monitoring on initiation of
drug use through late childhood. J Am Acad Child Adolesc Psychiatry, 35(1), 91100.
Chilcoat, H. D., Dishion, T. J., & Anthony, J. C. (1995). Parent monitoring and the
incidence of drug sampling in urban elementary school children. Am J Epidemiol,
141(1), 25-31.
Clayton, R. R., Cattarello, A. M., & Johnstone, B. M. (1996). The effectiveness of Drug
Abuse Resistance Education (project DARE): 5-year follow-up results. Prev Med,
25(3), 307-318.
Coley, R. L., & Hoffman, L. W. (1996). Relations of parental supervision and monitoring
to children's functioning in various contexts: Moderating effects of families and
neighborhoods. Journal of Applied Developmental Psychology, 17(1), 51-68.
de Graaf, R., Radovanovic, M., van Laar, M., Fairman, B., Degenhardt, L., AguilarGaxiola, S., et al. Early cannabis use and estimated risk of later onset of
depression spells: Epidemiologic evidence from the population-based World
Health Organization World Mental Health Survey Initiative. Am J Epidemiol,
172(2), 149-159.
Degenhardt, L., Chiu, W. T., Sampson, N., Kessler, R. C., Anthony, J. C., Angermeyer,
M., et al. (2008). Toward a global view of alcohol, tobacco, cannabis, and cocaine
use: findings from the WHO World Mental Health Surveys. PLoS Med, 5(7),
e141.
Dick, D. M., Latendresse, S. J., Lansford, J. E., Budde, J. P., Goate, A., Dodge, K. A., et
al. (2009). Role of GABRA2 in trajectories of externalizing behavior across
development and evidence of moderation by parental monitoring. Arch Gen
Psychiatry, 66(6), 649-657.
Dick, D. M., Viken, R., Purcell, S., Kaprio, J., Pulkkinen, L., & Rose, R. J. (2007).
Parental monitoring moderates the importance of genetic and environmental
influences on adolescent smoking. J Abnorm Psychol, 116(1), 213-218.
DiClemente, R. J., Wingood, G. M., Crosby, R., Sionean, C., Cobb, B. K., Harrington,
K., et al. (2001). Parental monitoring: association with adolescents' risk behaviors.
Pediatrics, 107(6), 1363-1368.

134

Dishion, T. J., & Kavanagh, K. (2000). A multilevel approach to family-centered
prevention in schools: process and outcome. Addict Behav, 25(6), 899-911.
Dishion, T. J., & McMahon, R. J. (1998). Parental monitoring and the prevention of child
and adolescent problem behavior: a conceptual and empirical formulation. Clin
Child Fam Psychol Rev, 1(1), 61-75
.
Fergusson, D. M., & Boden, J. M. (2008a). Cannabis use and adult ADHD symptoms.
Drug Alcohol Depend, 95(1-2), 90-96.
Fergusson, D. M., & Boden, J. M. (2008b). Cannabis use and later life outcomes.
Addiction, 103(6), 969-976; discussion 977-968.
Fergusson, D. M., & Horwood, L. J. (2000). Does cannabis use encourage other forms of
illicit drug use? Addiction, 95(4), 505-520.
Fergusson, D. M., Horwood, L. J., & Ridder, E. M. (2007). Conduct and attentional
problems in childhood and adolescence and later substance use, abuse and
dependence: results of a 25-year longitudinal study. Drug Alcohol Depend, 88
Suppl 1, S14-26.
Fergusson, D. M., Horwood, L. J., & Swain-Campbell, N. (2002). Cannabis use and
psychosocial adjustment in adolescence and young adulthood. Addiction, 97(9),
1123-1135.
Fergusson, D. M., Swain-Campbell, N. R., & Horwood, L. J. (2002). Deviant peer
affiliations, crime and substance use: a fixed effects regression analysis. J Abnorm
Child Psychol, 30(4), 419-430.
Flory, K., Milich, R., Lynam, D. R., Leukefeld, C., & Clayton, R. (2003). Relation
between childhood disruptive behavior disorders and substance use and
dependence symptoms in young adulthood: individuals with symptoms of
attention-deficit/hyperactivity disorder and conduct disorder are uniquely at risk.
Psychol Addict Behav, 17(2), 151-158.
Gfroerer, J. (1987). Correlation between drug use by teenagers and drug use by older
family members. Am J Drug Alcohol Abuse, 13(1-2), 95-108.
Gfroerer, J. C., Wu, L.-T., & Penne, M. A. (2002). Initiation of Marijuana Use: Trends,
Patterns, and Implications (Analytic Series: A-17, DHHS Publication No. SMA
02-3711). Rockville, MD: Substance Abuse and Mental Health Services
Administration, Office of Applied Studies.
Gillespie, N. A., Neale, M. C., & Kendler, K. S. (2009). Pathways to cannabis abuse: a
multi-stage model from cannabis availability, cannabis initiation and progression
to abuse. Addiction, 104(3), 430-438.
135

Grant, J. D., Scherrer, J. F., Lynskey, M. T., Lyons, M. J., Eisen, S. A., Tsuang, M. T., et
al. (2006). Adolescent alcohol use is a risk factor for adult alcohol and drug
dependence: evidence from a twin design. Psychol Med, 36(1), 109-118.
Guo, J., Hill, K. G., Hawkins, J. D., Catalano, R. F., & Abbott, R. D. (2002). A
developmental analysis of sociodemographic, family, and peer effects on
adolescent illicit drug initiation. J Am Acad Child Adolesc Psychiatry, 41(7), 838845.
.
Harrison, L. D. (1995). The validity of self-reported data on drug use. [Article]. Journal
of Drug Issues, 25(1), 91-111.
Hayatbakhsh, M. R., Mamun, A. A., Najman, J. M., O'Callaghan, M. J., Bor, W., & Alati,
R. (2008). Early childhood predictors of early substance use and substance use
disorders: prospective study. Aust N Z J Psychiatry, 42(8), 720-731.
Henquet, C., Krabbendam, L., Spauwen, J., Kaplan, C., Lieb, R., Wittchen, H. U., et al.
(2005). Prospective cohort study of cannabis use, predisposition for psychosis,
and psychotic symptoms in young people. BMJ, 330(7481), 11.
Hofler, M., Lieb, R., Perkonigg, A., Schuster, P., Sonntag, H., & Wittchen, H. U. (1999).
Covariates of cannabis use progression in a representative population sample of
adolescents: a prospective examination of vulnerability and risk factors.
Addiction, 94(11), 1679-1694.
Johnson, C., Drgon, T., Liu, Q. R., Zhang, P. W., Walther, D., Li, C. Y., et al. (2008).
Genome wide association for substance dependence: convergent results from
epidemiologic and research volunteer samples. BMC Med Genet, 9, 113.
Johnston, L. D., O'Malley, P. M., Bachman, J. G., & Schulenberg, J. E. (2009).
Monitoring the Future national survey results on drug use, 1975-2008. Volume I:
Secondary school students. Bethesda, MD: National Institute on Drug Abuse.
Kandel, D. (1975). Stages in adolescent involvement in drug use. Science, 190(4217),
912-914.
Kandel, D., & Faust, R. (1975). Sequence and stages in patterns of adolescent drug use.
Arch Gen Psychiatry, 32(7), 923-932.
Kandel, D., & Yamaguchi, K. (1993). From beer to crack: developmental patterns of drug
involvement. Am J Public Health, 83(6), 851-855.
Kandel, D. B. (1984). Marijuana users in young adulthood. Arch Gen Psychiatry, 41(2),
200-209.

136

Kandel, D. B., & Yamaguchi, K. (1985). Developmental patterns of the use of legal,
illegal, and medically prescribed psychotropic drugs from adolescence to young
adulthood. NIDA Res Monogr, 56, 193-235.
Kandel, D. B., Yamaguchi, K., & Chen, K. (1992). Stages of progression in drug
involvement from adolescence to adulthood: further evidence for the gateway
theory. J Stud Alcohol, 53(5), 447-457.
Kaplan, H. B., Martin, S. S., & Robbins, C. (1984). Pathways to adolescent drug use:
self-derogation, peer influence, weakening of social controls, and early substance
use. J Health Soc Behav, 25(3), 270-289.
Kellam, S. G., & Anthony, J. C. (1998). Targeting early antecedents to prevent tobacco
smoking: findings from an epidemiologically based randomized field trial. Am J
Public Health, 88(10), 1490-1495.
Kellam, S. G., Brown, C. H., Poduska, J. M., Ialongo, N. S., Wang, W., Toyinbo, P., et
al. (2008). Effects of a universal classroom behavior management program in first
and second grades on young adult behavioral, psychiatric, and social outcomes.
Drug Alcohol Depend, 95 Suppl 1, S5-S28.
Kellam, S. G., Reid, J., & Balster, R. L. (2008). Effects of a universal classroom behavior
program in first and second grades on young adult problem outcomes. Drug
Alcohol Depend, 95 Suppl 1, S1-4.
Kellam, S. G., Werthamer-Larsson, L., Dolan, L. J., Brown, C. H., Mayer, L. S., Rebok,
G. W., et al. (1991). Developmental epidemiologically based preventive trials:
baseline modeling of early target behaviors and depressive symptoms. Am J
Community Psychol, 19(4), 563-584.
Kendler, K. S., Karkowski, L. M., Neale, M. C., & Prescott, C. A. (2000). Illicit
psychoactive substance use, heavy use, abuse, and dependence in a US
population-based sample of male twins. Arch Gen Psychiatry, 57(3), 261-269.
Kendler, K. S., & Prescott, C. A. (1998). Cannabis use, abuse, and dependence in a
population-based sample of female twins. Am J Psychiatry, 155(8), 1016-1022.
Kenny, D. A. (2009). Mediation. Retrieved November 1, 2010, from
http://davidakenny.net/cm/mediate.htm
Kosterman, R., Hawkins, J. D., Guo, J., Catalano, R. F., & Abbott, R. D. (2000). The
dynamics of alcohol and marijuana initiation: patterns and predictors of first use
in adolescence. Am J Public Health, 90(3), 360-366.
Kumpfer, K. L., & Alvarado, R. (2003). Family-strengthening approaches for the
prevention of youth problem behaviors. Am Psychol, 58(6-7), 457-465.
137

Kurdek, L. A., Fine, M. A., & Sinclair, R. J. (1995). School Adjustment in Sixth Graders:
Parenting Transitions, Family Climate, and Peer Norm Effects. Child
Development, 66(2), 430-445.
Lloyd, J. J., & Anthony, J. C. (2003). Hanging out with the wrong crowd: how much
difference can parents make in an urban environment? J Urban Health, 80(3),
383-399.
Lynam, D. R., Milich, R., Zimmerman, R., Novak, S. P., Logan, T. K., Martin, C., et al.
(1999). Project DARE: no effects at 10-year follow-up. J Consult Clin Psychol,
67(4), 590-593.
Lynskey, M., & Hall, W. (2000). The effects of adolescent cannabis use on educational
attainment: a review. Addiction, 95(11), 1621-1630.
Lynskey, M. T., Heath, A. C., Bucholz, K. K., Slutske, W. S., Madden, P. A., Nelson, E.
C., et al. (2003). Escalation of drug use in early-onset cannabis users vs co-twin
controls. JAMA, 289(4), 427-433.
Lynskey, M. T., Heath, A. C., Nelson, E. C., Bucholz, K. K., Madden, P. A., Slutske, W.
S., et al. (2002). Genetic and environmental contributions to cannabis dependence
in a national young adult twin sample. Psychol Med, 32(2), 195-207.
Magura, S., & Kang, S. Y. (1996). Validity of self-reported drug use in high risk
populations: a meta-analytical review. Subst Use Misuse, 31(9), 1131-1153.
Martins, S. S., Storr, C. L., Alexandre, P. K., & Chilcoat, H. D. (2008). Adolescent
ecstasy and other drug use in the National Survey of Parents and Youth: the role
of sensation-seeking, parental monitoring and peer's drug use. Addict Behav,
33(7), 919-933.
Merikangas, K. R., Li, J. J., Stipelman, B., Yu, K., Fucito, L., Swendsen, J., et al. (2009).
The familial aggregation of cannabis use disorders. Addiction, 104(4), 622-629.
Merikangas, K. R., Stolar, M., Stevens, D. E., Goulet, J., Preisig, M. A., Fenton, B., et al.
(1998). Familial transmission of substance use disorders. Arch Gen Psychiatry,
55(11), 973-979.
Miles, D. R., van den Bree, M. B., Gupman, A. E., Newlin, D. B., Glantz, M. D., &
Pickens, R. W. (2001). A twin study on sensation seeking, risk taking behavior
and marijuana use. Drug Alcohol Depend, 62(1), 57-68.
Molina, B. S., & Pelham, W. E., Jr. (2003). Childhood predictors of adolescent substance
use in a longitudinal study of children with ADHD. J Abnorm Psychol, 112(3),
497-507.
138

Monahan, K. C., Steinberg, L., & Cauffman, E. (2009). Affiliation with antisocial peers,
susceptibility to peer influence, and antisocial behavior during the transition to
adulthood. Dev Psychol, 45(6), 1520-1530.
Morral, A. R., McCaffrey, D. F., & Paddock, S. M. (2002). Reassessing the marijuana
gateway effect. Addiction, 97(12), 1493-1504.
Morral, A. R., McCafrey, D. F., & Paddock, S. M. (2002). Evidence does not favor
marijuana gateway effects over a common-factor interpretation of drug use
initiation: responses to Anthony, Kenkel & Mathios and Lynskey. Addiction,
97(12), 1509-1510.
Morrison, P. D., Zois, V., McKeown, D. A., Lee, T. D., Holt, D. W., Powell, J. F., et al.
(2009). The acute effects of synthetic intravenous Delta9-tetrahydrocannabinol on
psychosis, mood and cognitive functioning. Psychol Med, 39(10), 1607-1616.
Muthén, L. K., & Muthén, B. O. (1998-2010). Mplus User's Guide (Sixth Edition ed.).
Los Angeles, CA: Muthén & Muthén.
Patterson, G. R., DeBaryshe, B. D., & Ramsey, E. (1989). A developmental perspective
on antisocial behavior. Am Psychol, 44(2), 329-335.
Reboussin, B. A., Hubbard, S., & Ialongo, N. S. (2007). Marijuana use patterns among
African-American middle-school students: a longitudinal latent class regression
analysis. Drug Alcohol Depend, 90(1), 12-24.
Rios-Bedoya, C. F., Wilcox, H. C., Piazza, M., & Anthony, J. C. (2008). Children taking
risks: the association with cocaine and other drug use by young adulthood. Addict
Behav, 33(9), 1154-1161.
Rosenberg, M. F., & Anthony, J. C. (2001a). Aggressive behavior and opportunities to
purchase drugs. Drug Alcohol Depend, 63(3), 245-252.
Rosenberg, M. F., & Anthony, J. C. (2001b). Early clinical manifestations of cannabis
dependence in a community sample. Drug Alcohol Depend, 64(2), 123-131.
Royston, P., & Altman, D. G. (1994). Regression Using Fractional Polynomials of
Continuous Covariates - Parsimonious Parametric Modeling. Applied StatisticsJournal of the Royal Statistical Society Series C, 43(3), 429-467.
Spoth, R., Guyll, M., & Shin, C. (2009). Universal intervention as a protective shield
against exposure to substance use: long-term outcomes and public health
significance. Am J Public Health, 99(11), 2026-2033.

139

Stallings, M. C., Corley, R. P., Dennehey, B., Hewitt, J. K., Krauter, K. S., Lessem, J. M.,
et al. (2005). A genome-wide search for quantitative trait Loci that influence
antisocial drug dependence in adolescence. Arch Gen Psychiatry, 62(9), 10421051.
Stallings, M. C., Corley, R. P., Hewitt, J. K., Krauter, K. S., Lessem, J. M., Mikulich, S.
K., et al. (2003). A genome-wide search for quantitative trait loci influencing
substance dependence vulnerability in adolescence. Drug Alcohol Depend, 70(3),
295-307.
StataCorp. (2009). Stata Statistical Software: Release 11. College Station, TX: StataCorp
LP.
Steinberg, L., & Monahan, K. C. (2007). Age differences in resistance to peer influence.
Dev Psychol, 43(6), 1531-1543.
Substance Abuse and Mental Health Service Administration. (2009). Results from the
2008 National Survey on Drug Use and Health: National Findings Rockville,
MD.
Tang, Z., & Orwin, R. G. (2009). Marijuana initiation among American youth and its
risks as dynamic processes: prospective findings from a national longitudinal
study. Subst Use Misuse, 44(2), 195-211.
Tien, A. Y., & Anthony, J. C. (1990). Epidemiological analysis of alcohol and drug use
as risk factors for psychotic experiences. J Nerv Ment Dis, 178(8), 473-480.
UCLA. How can I perform mediation with binary variables? Retrieved November 1,
2010, from http://www.ats.ucla.edu/stat/stata/faq/binary_mediation.htm
Uhl, G. R., Liu, Q. R., Walther, D., Hess, J., & Naiman, D. (2001). Polysubstance abusevulnerability genes: genome scans for association, using 1,004 subjects and 1,494
single-nucleotide polymorphisms. Am J Hum Genet, 69(6), 1290-1300.
UNODC. (2010). World Drug Report 2010: (United Nations Publication, Sales No.
E.10.XI.13).
van den Bree, M. B., & Pickworth, W. B. (2005). Risk factors predicting changes in
marijuana involvement in teenagers. Arch Gen Psychiatry, 62(3), 311-319.
Van Etten, M. L., & Anthony, J. C. (1999). Comparative epidemiology of initial drug
opportunities and transitions to first use: marijuana, cocaine, hallucinogens and
heroin. Drug and Alcohol Dependence, 54(2), 117-125.

140

van Os, J., Bak, M., Hanssen, M., Bijl, R. V., de Graaf, R., & Verdoux, H. (2002).
Cannabis use and psychosis: a longitudinal population-based study. Am J
Epidemiol, 156(4), 319-327.
von Sydow, K., Lieb, R., Pfister, H., Hofler, M., & Wittchen, H. U. (2002). What predicts
incident use of cannabis and progression to abuse and dependence? A 4-year
prospective examination of risk factors in a community sample of adolescents and
young adults. Drug Alcohol Depend, 68(1), 49-64.
Wagner, F. A., & Anthony, J. C. (2002a). From first drug use to drug dependence;
developmental periods of risk for dependence upon marijuana, cocaine, and
alcohol. Neuropsychopharmacology, 26(4), 479-488.
Wagner, F. A., & Anthony, J. C. (2002b). Into the world of illegal drug use: exposure
opportunity and other mechanisms linking the use of alcohol, tobacco, marijuana,
and cocaine. Am J Epidemiol, 155(10), 918-925.
Wells, J. E., Degenhardt, L., Bohnert, K. M., Anthony, J. C., & Scott, K. M. (2009).
Geographical clustering of cannabis use: results from the New Zealand Mental
Health Survey 2003-2004. Drug Alcohol Depend, 99(1-3), 309-316.
Wilcox, H. C., Storr, C. L., Benoit, M. Z., & Anthony, J. C. (2005). Ethnicity as manifest
in drug-specific vocabulary and subsequent risk of starting cannabis use in early
adolescence. Ethn Dis, 15(3), 485-491.
Winters, K. C., & Lee, C. Y. (2008). Likelihood of developing an alcohol and cannabis
use disorder during youth: association with recent use and age. Drug Alcohol
Depend, 92(1-3), 239-247.
Yamaguchi, K., & Kandel, D. B. (1984a). Patterns of drug use from adolescence to
young adulthood: II. Sequences of progression. Am J Public Health, 74(7), 668672.
Yamaguchi, K., & Kandel, D. B. (1984b). Patterns of drug use from adolescence to
young adulthood: III. Predictors of progression. Am J Public Health, 74(7), 673681.
Zeger, S. L., Liang, K. Y., & Albert, P. S. (1988). Models for longitudinal data: a
generalized estimating equation approach. Biometrics, 44(4), 1049-1060.
Zou, G. (2004). A modified poisson regression approach to prospective studies with
binary data. Am J Epidemiol, 159(7), 702-706.

141