Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Antecedents of marijuana initiation
(USC Thesis Other)
Antecedents of marijuana initiation
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
ANTECEDENTS MARIJUANA INITIATION
Antecedents of Marijuana Initiation
by
Nicholas J. Jackson
A Thesis Submitted in
Partial Fulfillment of the
Requirements for the Degree of
Master of Arts
in Psychology
at
The University of Southern California
December 2014
ANTECEDENTS MARIJUANA INITIATION
Table of Contents
ABSTRACT……………………………………………………………………………..... 6
INTRODUCTION………………………………………………………………………... 7-8
METHODS
Participants & Procedure………………………………………………………… 9
Measures…………………………………………………………………………. 9-10
Analysis…………………………………………………………………………... 10-13
RESULTS
Description of Sample……………………………………………………………. 13-14
Attrition…………………………………………………………………………... 14
Identifying Predictors of Marijuana Initiation Age Groups……………………… 15
Predictors of Initiation in Early Adolescence……………………………….…… 15-17
Predictors of Initiation in Mid-Adolescence……………………………………... 17-18
Predictors of Initiation in Late Adolescence……………………………………... 18-19
Gradient Boosted Modeling……………………………………………………… 19-21
DISUCSSION
Impact of Attrition on Marijuana Outcomes……………………………………... 21-22
Exploratory Analysis and Methods………………………………………………. 22-24
Predictors of Initiation Age Group………………………………………………. 24-26
Predictors of Ever Use…………………………………………………………… 27-30
Strengths and Limitations………………………………………………………... 30
Future Directions………………………………………………………………… 30
ANTECEDENTS MARIJUANA INITIATION
ACKNOWLEDGEMENTS………………………………………………………………. 31
REFERENCES……………………………………………………………………………. 32-39
METHODS SUPPLEMENT……………………………………………………………… 52-55
APPENDIX
Neighborhood Characteristics Scale…………………………………………...… 60
Perceived Stress Scale……………………………………………………………. 61
Childhood Friendship Questionnaire Scale……………………………………… 62-63
Parent-to-Child Affect Scale……………………………………………………... 64
ANTECEDENTS MARIJUANA INITIATION
List of Tables
Table Page
1 Distributions of Marijuana Outcomes…………………………………..….….…. 39
2 Bivariate Associations with Marijuana Initiation Age Groups………….….….… 40-41
3.1 MLoR of Socio-demographic predictors-Early Adolescence vs Never Initiated.. 42
3.2 MLoR of Socio-demographic predictors-Mid-Adolescence vs Never Initiated…. 43
3.3 MLoR of Socio-demographic predictors-Late Adolescence vs Never Initiated…. 44
4.1 MLoR of Predictors-Early Adolescence vs Never Initiated…………..…..…..….. 45
4.2 MLoR of Predictors-Mid-Adolescence vs Never Initiated…………….…..…..…. 46
4.3 MLoR of Predictors-Late Adolescence vs Never Initiated………………..…..…. 46
S1 Attrition Analyses……………………………………………………................... 56-57
S2 Relative Influence of Predictors in GBM……………………………................... 58
ANTECEDENTS MARIJUANA INITIATION
List of Figures
Figure Page
1 Multivariable MLoR……………………………………………………….….… 48
2 Cross-Validation to Determine Interaction ……………………………….….… 48
3 Cross-Validation to Determine Tree Size…………………………...…….….… 49
4 Relative Influence of Predictors of “Ever Use” from GBM…………...….….… 49
5 ROC Curve of GBM Predictions……………………………………….….….… 50
S1 Patterns of Missing Data & Using Reasonable Inference…………….……....… 59
S2 GBM Predictors Association with Ever Use……………………………………. 59
ANTECEDENTS MARIJUANA INITIATION
6
Abstract
Marijuana use among adolescents has been increasing and is associated with loss in cognitive
abilities, risky behavior, and later substance abuse and dependence. Specifically, early initiation
is associated with long-term negative outcomes. Data from The University of Southern
California (USC) Risk Factors for Antisocial Behavior (RFAB) twin study were used to examine
the psychosocial antecedents of marijuana initiation in adolescence. Participants were assessed in
Wave 1 (ages 9-10) for risk and resiliency factors prior to ever having used marijuana. Marijuana
use was assessed every 2-3 years, with the participants currently 19-20 years old (Wave 5-
ongoing). The sample was randomly split such that one twin was assigned to an exploratory
dataset and the co-twin to a testing (or confirmation) dataset. The exploratory dataset was used
for model building and the testing dataset was used to determine the robustness or accuracy of
the model. Wave 1 predictors were used in multinomial logistic regression models to predict
initiation age groups. Gradient Boosted Modeling was used to identify influential predictors of
ever having used marijuana. Predictors of initiation were related to friendships and bully
victimization, sociodemographics, parental stress, and the parent-child relationship.
Keywords: marijuana initiation, twin study, substance use, adolescence, statistical
learning, machine learning, gradient boosting
ANTECEDENTS MARIJUANA INITIATION
7
Antecedents of Marijuana Initiation
Marijuana is the fourth most commonly used drug behind caffeine, tobacco, and alcohol
in the United States with a lifetime prevalence of 50% (NSDUH, 2012; Shi, 2014). Of illicit
drugs, marijuana is the most commonly used with 18.9 million Americans having reported use in
the past 30 days (SAMHSA 2012; NSDUH, 2012; Popovici, French, Pacula, Maclean, &
Antonaccio, 2014). The consequences of this use have broad effects on the individual and
society. Comorbidity with mental disorders such as depression, anxiety, and anti-social
personality can have a lasting negative impact on the individual (Chen, Wagner, & Anthony,
2002; Macleaod et al., 2004; Chassin, Pitts, DeLucia, & Todd, 1999; Kokkevi, Richardson,
Florescu, Kuzman, & Stergar, 2007). These individual factors can have a high societal cost
through increased unemployment, absenteeism, decreased productivity, and increased crime and
rates of incarceration (DeSimone, 2002; Fergusson & Bowden, 2008; Pedersen & Skardhamar,
2010; Pacula & Kilmer, 2003). Particularly, use in adolescence, when the brain is still
developing, has shown implications for future neurocognitive decline (Meier et al., 2012;
Lisdahl, Wright, Medina-Kirchner, 2014), psychiatric disorders (Arseneault, Cannon, Witton, &
Murray, 2004; Semple, McIntosh, & Lawrie, 2005; Moore et al, 2007), impaired scholastic
achievement (Bell, Weschler, & Johnson, 1997; Finnell & Jones, 1975; Lyskey & Hall, 2000),
diminished lifetime earnings (Ringel, Ellickson, & Collins, 2006), and increased rates of
incarceration (Pedersen & Skardhamar, 2010) . Given that 18.7% of late adolescents (ages 18-
25) in the US have used marijuana in the past month (Popovici et al., 2014), the potential impact
is non-trivial.
The origins of substance use disorders are complex and involve the interactions of
genetic vulnerability, environment (family, school, peer, neighborhood, culture),
ANTECEDENTS MARIJUANA INITIATION
8
sociodemographic characteristics, and personality; the relative influences of which may change
across development. While much has been studied on the origins of alcohol use disorders,
research surrounding marijuana use and dependency is relatively scarce.
From the alcohol literature, we know that age at initiation is one of the most important
factors in determining the progression and outcome for an alcohol use disorder (Sher, Grekin, &
Williams, 2005; DeWitt, Adlaf, Offord, & Ogborne, 2000). Similar findings have been found for
marijuana initiation (Ellickson, Tucker, Klein, & Saner, 2004; Brook, Adams, Balka, & Johnson,
2002; Flory, Lynam, Milich, Leukefeld, & Clayton, 2004 ), though much more research needs to
be done in this area. Research on predictors of marijuana initiation is especially salient now
given recent changes in the social and legal attitudes towards marijuana use in the United States
(Colorado Const. amed. LXIV; Remnick, 2014). While many studies have reported on the
correlates of marijuana use, few have the ability to examine true predictors, where reporting of
the exposure temporally precedes initiation. This work will expand upon previous studies that
have examined longitudinal antecedents of marijuana initiation (eg. Siegel et al, 2013; Ellickson
et al., 2004; Jessor ,1976; Kosterman, Hawkins, Guo, Catalano, & Abbott, 2000; Brook, Balka,
& Whiteman, 1999)
The present study builds upon the previous literature regarding marijuana initiation by
examining how a large (62 variables) and diverse set of predictors (sociodemographic, substance
use, family & peer relationships, parental psychosocial, and personality) from late childhood (9-
10 years old) can be used to predict marijuana initiation in adolescence. This study makes use of
classical statistical models and modern algorithmic based approaches in a twin sample to identify
the antecedents of marijuana initiation.
ANTECEDENTS MARIJUANA INITIATION
9
Method
Participants & Procedure
This investigation was part of a larger study of twins in the University of Southern
California Risk Factors for Antisocial Behavior (RFAB) study. RFAB in an ongoing longitudinal
study of 1555 monozygotic and dizygotic twins recruited from the greater Los Angeles area
school districts. Participants were between 9 and 10 years old (mean=9.6) at the start of the first
wave and were assessed on five occasions (~every 3 years) for psychosocial, behavioral,
personality, and socio-demographic risk factors of substance use. Participants are currently in
Wave 5 (ages 19-20). A more complete description of the RFAB study can be found in Baker,
Barton, and Raine (2002).
Measures
Marijuana use. Marijuana use was assessed at each wave by self-report to the following
question: “Have you ever tried marijuana?”. Responses options were Yes or No. Missing data on
marijuana use was filled in for each participant using a reasonable inference, such that No
responses in a later wave were used to infer a No response in a former wave. Similarly, a Yes
response at an earlier time-point was used to infer Yes responses in later waves. From these
responses, two outcome variables pertaining to marijuana use were created: Initiation Age Group
and Ever Use. Initiation Age Groups were defined as Never initiated, Early Adolescence (by
Wave 3: 14-15 Years), Mid-Adolescence (Wave 4: 16-18 Years), and Late Adolescence (Wave
5: 19-20 Years). Only subjects with complete data after inferring usage were able to be classified
into these groups. The Ever Use variable was defined based on a subject ever reporting having
used marijuana at any wave. Data at Wave 5 was required for creation of this variable.
ANTECEDENTS MARIJUANA INITIATION
10
Supplementary Figure 1 provides examples of how missing values were inferred and the effects
of this inference on the two outcomes.
Predictors of marijuana use. Sixty-two variables collected at Wave 1 from the twin and
parent, prior to any twin reporting marijuana use, were used to predict future marijuana use.
These predictors were chosen based upon a literature search for predictors of substance use
initiation (alcohol, drugs, and marijuana). These measures were: Socio-demographic variables
such as age at Wave 1, gender, race, Zygosity, socioeconomic status (based on Hollingshead
ratings), handedness, and neighborhood characteristics score; Substance-Use variables such as
twin-reported cigarette or alcohol use, peer cigarette, alcohol, marijuana, or drug use, and
mother’s cigarette, alcohol, or drug use during pregnancy; Family & Peer Relationship variables
such as parent and twin-report on positive/negative parent-to-child affect, parent’s relationship
with their partner, and variables pertaining to twin friendships and bully victimization; Parental
Psychosocial variables such as perceived stress and prenatal support; and Personality variables
such as factors from the Junior Temperament and Character Inventory (JTCI) (Luby, Svrakic,
McCallum, Przybeck, & Cloninger, 1999) and the Childhood Psychopathy Scale (CPS) (Lynam,
1997). A more detailed description of these measures are provided in the supplement.
Analysis
To facilitate the analytic strategy, the sample was randomly split such that one twin was
assigned to an exploratory dataset and the co-twin to a testing (or confirmation) dataset. The
general strategy was to use the exploratory dataset for model building and then use the testing
dataset to determine the robustness of the model. This approach was taken using two techniques
applied to this data 1) multinomial logistic regression and 2) Gradient Boosted Modeling.
ANTECEDENTS MARIJUANA INITIATION
11
Attrition. Four levels of attrition with respect to the outcome were determined for this
longitudinal study. Complete attrition refers to participants that took part only in Wave 1 of the
study. Partial attrition refers to participants that participated in more than just Wave 1, but still do
not have complete data. Partial attrition with inference refers to participants that have complete
data, but only because of the ability to make inferences from the patterns of their responses per
Figure S1 in the supplement. Complete cases were those participants that participated in all five
waves. A bivariate assessment of each predictor of marijuana use with the attrition variable was
examined in the exploratory dataset using ANOVA (for continuous predictors) or a Fisher’s
Exact Chi-square (for discrete predictors). Results are presented in supplementary Table S1.
Predictors with associations < 0.05 were used for adjustment in multinomial logistic regression
models (described below).
Multinomial logistic regression. The marijuana initiation age group outcome was
analyzed using a multinomial logistic regression (MLoR) model comparing each initiation age
group to those that never initiated. In the exploratory dataset, bivariate assessment (ANOVA and
Fisher’s exact) of each predictor with initiation was used to determine which variables should be
of further interest for model building. Predictors with bivariate associations with p<0.20 were
used to build a series of hierarchical regressions that were unadjusted (model 1), adjusted for
socio-demographic information in model 2 (age, race, gender, Zygosity, and neighborhood
characteristics score) and for variables associated with attrition (see Table S1) in model 3. Use of
a relaxed significance level for model development is common due to the fact that weak bivariate
associations may become significant after covariate adjustment (Hosmer & Lemeshow, 1989).
For each of the models built in the exploratory dataset with statistically significant associations
(p<0.05), a cross-sample (testing dataset) estimate of effect replication was conducted in the
ANTECEDENTS MARIJUANA INITIATION
12
testing dataset. The estimate of effect replication was created by running the same MLoR on
1000 bootstrap samples of the testing data and determining the proportion of times the effect
detected in the exploratory data was within the 95% confidence interval on the bootstrap sample
and also was statistically significant at p<0.05. This can be considered as analogous to the
empirical statistical power for each of these effects in a novel sample (albeit genetically related).
Analyses were conducted using Stata Version 12.1 StataCorp, LP (College Station, TX).
Gradient boosted modeling. Statistical (or Machine) learning has received much hype in
the literature in recent years as part of the principal set of methodologies for analyzing “big
data”. An asset of these techniques is the ability to uncover non-linear and high-order
interactions that provide the maximal predictive accuracy for the outcome given the dataset.
Gradient Boosted Modeling (GBM) is one such technique that combines the simplicity of
decision trees with a high level of predictive accuracy. This is achieved by growing a large
number of decision trees, with each tree fit sequentially such that each subsequent tree is fit to
shrunken (down weighted) residuals from the previous tree (Friedman, 2001; Friedman, 2002).
The number of trees, the learning (or shrinkage) rate, and number of terminal nodes (interaction
depth) for each tree can be set to provide the optimal prediction. In these analyses, a learning rate
(shrinkage parameter) of 0.001 was used per Friedman (2001) suggesting better performance
with slower learning rates. In the exploratory dataset, 3000 gradient boosted trees were fit to the
Ever Use marijuana dichotomous outcome using all 62 predictors. Ten-fold cross-validation was
used to determine the optimal (minimized generalization error) tree number and interaction
depth. Predictors in the GBM were evaluated based upon the measure of relative influence,
which was computed as the average across all tress of the empirical improvement of the model
from splitting a decision tree on the predictor. The GBM was conducted using the gbm package
ANTECEDENTS MARIJUANA INITIATION
13
(Ridgeway, 2013) in R version 2.14.0. Once fit to the exploratory dataset, predicted probability
of Ever Use was applied in the testing dataset with classification accuracy assessed.
Classification accuracy was evaluated by first determining the operating point on a Receiver
Operating Curve (ROC) of the exploratory data. The operating point was defined to be the point
of predicted probability that maximized sensitivity and specificity on the ROC curve assuming
equal costs for misclassification. Mathematically this reduces to finding the observed predicted
probability (pr) that minimizes the cost function: [( ) ( ) (
)]. Once obtained, this operating point is used to classify participants in the testing
dataset as never users or users of marijuana. This classification is compared to the truth
contained within the Ever Use variable to determine model accuracy. An alternate measure of
ROC fit, Area Under the Curve (AUC) was also used. AUC represents the probability that a
randomly selected participant who has used marijuana would have a greater predicted probability
of use in the model compared to a never user. A value of 0.5 represents classification at chance
levels. A non-parametric bootstrap was used to determine AUC and 95% confidence intervals.
Results
Description of Sample
Fifteen hundred fifty-five twins from 778 families started at Wave 1 (age=9.6 ± 0.59) of
this ongoing longitudinal study. The sample was 50% Female with 43% of the twins being
monozygotic. Racially, the sample was primarily Hispanic (35%) and Caucasian (30%) with
smaller numbers of multiracial (18%), African-American (13%), and Asian (4%) participants.
These numbers are largely consistent with the ethnoracial makeup of the greater Los Angeles
area at the time of Wave 1 enrollment (January 2001) with the exception of Asian participants
who are underrepresented in this sample (US Census Bureau, 2002). After randomly splitting
ANTECEDENTS MARIJUANA INITIATION
14
the sample, 778 twins were assigned to the exploratory dataset. Of these, 432 participants had
information regarding marijuana Ever Use and 351 had data available for the Initiation Age
Groups outcome. The 777 twins assigned to the testing dataset had N ‘s of 414 for Ever Use and
335 for the Initiation Age Groups outcome. Distributions of these outcomes were similar
between the two samples (Table 1) with a pooled average of 56.5% of the sample having used
marijuana. The most common age to initiated marijuana use was in mid-adolescence (16-18
years). Demographics for those with available data are presented in supplementary Table S1.
Attrition
The attrition rate for the study with regard to the outcomes was 22.4% over the ten years
of this longitudinal study, indicating that a majority of subjects participated in more than just the
first Wave. Additionally, this number is an overestimation due to the fact that Wave 5 data
collection is ongoing and these analyses require complete data for the Initiation Age Group
outcome. Table S1 in the supplement provides the results of the attrition analysis with the
exploratory dataset. Approximately 22% of the subjects had complete attrition (only participated
in Wave 1), 33% had partial attrition (participated in more than just Wave 1), 24% had partial
attrition but were able to have the marijuana outcomes inferred from their previous responses
(per Figure S1), and 21% were complete cases (data at all 5 waves). Predictors of marijuana
initiation that were associated (p<0.05) with attrition were younger age at Wave 1,
monozygosity, left handedness, peers having used marijuana, parents not wanting their
pregnancy of the twins, higher levels of psychopathy, higher scores on the JTCI novelty seeking
and harm avoidance scales, lower scores on the JTCI self-directed and cooperative scales, being
bullied in their neighborhood, and having their caregiver report greater amounts of stress. These
variables were used in fully adjusted models of the Initiation Age Group outcome.
ANTECEDENTS MARIJUANA INITIATION
15
Identifying Predictors of Marijuana Initiation Age Groups
Table 2 shows the results of the bivariate analysis of the predictors with the marijuana
initiation age groups outcome in the exploratory dataset. Associations with p<0.20 were selected
for further multivariable modeling. In instances where two variables from the same
questionnaire were associated, the variables selected for multivariable modeling were chosen
based upon 1) having a lower p value if the variables were collinear (eg. Overall happiness in
relationship vs amount of effort partner puts in relationship); 2) being superordinate to the other
variables from the same questionnaire (eg. using overall friendship scale rather than the
component initiator and receiver scales); 3) having adequate cell size across the initiation age
groups (eg. Choosing twin sip of alcohol vs puff of cig and not using the PDI).
Predictors of Initiation in Early Adolescence
Results for predictors of initiation in early adolescence are displayed in Tables 3.1 & 4.1.
Figure 1 provides a graphic of the effect sizes and precision for the non-sociodemographic
predictors, presented in the order in which they appear in the tables.
Socio-demographic. Variables such as wave 1 age, Zygosity, and race were significantly
associated with marijuana initiation in early adolescence in models adjusting for other socio-
demographic variables and variables associated with attrition (Table 3.1). Older age at Wave 1
was found to confer an odds of initiation that was 2.21 times greater compared to participants
that were 1 year younger at the start; however this effect was only replicated 14% of the time in
the testing data, indicating this result is not robust within the sample. Asian ethnoracial
background was protective for early initiation compared to Caucasians (Odds Ratio=0.063),
though similarly this result was not robust (0% effect replication). Monozygotic twin were found
to have a 2-3 fold increased odds of initiation compared to dizygotic twins in the unadjusted
ANTECEDENTS MARIJUANA INITIATION
16
Model 1 (OR=2.67), socio-demographic adjusted Model 2 (OR=2.48), and fully adjusted Model
3 (OR=3.17). Effect replication for Zygosity was low, though greater than 50% for Model 1
(57.9%) and Model 3 (55.4%) indicating the result replicates for a majority of the bootstrap
samples.
Substance Use. Twins who reported having a sip of alcohol at Wave 1 had odds of early
initiation that were 2-3 times greater than those who had not had a sip of alcohol (Table 4.1).
While unadjusted results were robust (OR=2.53, 58.4% replication), after adjusting for other
potential confounds, the replication rates dropped dramatically (Model 2 OR=2.53, 24.4%;
Model 3 OR=3.83, 17.4%). Peer alcohol use was not associated with initiation in early
adolescence.
Family & Peer Relationships. The caregivers overall happiness in their romantic
relationship was not predictive of initiation. The twin’s relationship with their caregiver was
associated in fully adjusted models (Model 3) such that parent report of higher negative affect
was associated with 2.27 times odds of early initiation, though this result was not robust (16.8%
effect replication). Overall levels of friendship, bully victimization, and number of close friends
were modeled together to predict initiation in early adolescence. Reporting higher levels of
friendship/social interactions conferred a 2-3 fold risk of initiation in all Models (M1=2.15,
M2=2.53, M3=3.12) with these effect sizes being robust when examined in the testing dataset (%
effect replication 71.6%, 80.5%, 80.1%) . Bully victimization conferred an almost two-fold risk
(OR=1.87, 56.5%) in the unadjusted model, however this result showed poor replicability after
socio-demographic adjustment (OR=1.77, 39.5%). The number of close friends was not
significantly associated with initiation in early adolescence.
ANTECEDENTS MARIJUANA INITIATION
17
Parental Psychosocial. Higher levels of perceived stress by the caregiver were weakly
associated with initiation in the unadjusted model, with poor replication (OR=1.04, 16.6%).
These results were not significant after adjustment. Prenatal support variables were modeled
together to predict initiation. Having planned the pregnancy and having a supportive family were
both protective of initiation in early adolescence. Having planned the pregnancy resulted in odds
that were 1.5-6.5 times lower for initiation than those who did not plan their pregnancy
(M1=0.37, M2=0.22, M3=0.15) though the effect replication was only strong in for Model 2
(M1=29.8%, 74.3%, 51.0%). Having a supportive family was protective (odds ~1.5 times lower)
in unadjusted (OR=0.60, 58.1%) and socio-demographic adjusted models (OR=0.62, 48.2%),
though only the unadjusted model had replicability >50%. The happiness of the partner about the
pregnancy was not predictive of initiation in early adolescence.
Predictors of Initiation in Mid-Adolescence
Results for predictors of initiation in mid-adolescence are displayed in Tables 3.2 & 4.2.
Figure 1 provides a graphic of the effect sizes and precision for the non-sociodemographic
predictors, presented in the order in which they appear in the tables.
Socio-demographic. Female gender was protective against initiation in mid-adolescence
for the unadjusted (OR=0.54) and fully adjusted models (OR=0.38) such that females were 1.5-
2.5 times less likely to initiate compared to males. Effect replication for both models was low at
10.7% and 18.5% respectively (Table 3.2). No other socio-demographic variables were
significantly related to initiation.
Substance Use. Twins alcohol use was not associated with initiation in mid-adolescence.
Peer alcohol use was associated with a 7-9 fold increase in the odds of initiation in all models
(M1=9.21, M2=7.93, M3=7.31) with effect replication < 1% (Table 4.2).
ANTECEDENTS MARIJUANA INITIATION
18
Family & Peer Relationships. The caregivers overall happiness in their romantic
relationship was not predictive of initiation. The twin’s relationship with their caregiver was
associated in the unadjusted and socio-demographic adjusted models such that parent report of
higher negative affect was associated with 1.91 and 2.13 times odds of initiation respectively,
though this result was not robust (< 25% effect replication). Reporting higher levels of
friendship/social interactions resulted in a 1.5 fold risk of initiation in all Models (M1=1.45,
M2=1.54, M3=1.69) with these effect sizes being moderately robust when examined in the
testing dataset (% effect replication 60.6%, 62.4%, 61.1%). Bully victimization and the number
of close friends was not significantly associated with initiation in mid-adolescence.
Parental Psychosocial. Having planned the pregnancy was protective of initiation in
mid-adolescence in the demographic adjusted model such that having planned the pregnancy
resulted in odds that were 2.3 times lower for initiation than those who did not plan their
pregnancy (OR=0.43,68.9%). This effect had high replication, though was not significant in the
fully adjusted model. No other prenatal support variables were significant. Caregiver perceived
stress was not significantly associated with initiation.
Predictors of Initiation in Late Adolescence
Results for predictors of initiation in late adolescence are displayed in Tables 3.3 & 4.3.
Figure 1 provides a graphic of the effect sizes and precision for the non-sociodemographic
predictors, presented in the order in which they appear in the tables.
Socio-demographic. Being a monozygotic twin was a risk factor for initiation in late
adolescence in the unadjusted model (OR=2.84, 29.6%) with low replicability (Table 3.3). No
other socio-demographic variables were significantly related to initiation.
ANTECEDENTS MARIJUANA INITIATION
19
Substance Use. Twin and peer alcohol use was not associated with initiation in late
adolescence (Table 4.3).
Family & Peer Relationships. The caregivers overall happiness in their romantic
relationship was predictive of a 1.5-2.5 fold increase in the odds of initiation such that greater
unhappiness resulted in greater risk of initiation in Late Adolescence (M1=1.87, M2=2.34,
M3=2.59). However, these results has low replicability (<25% effect replication). Reporting
higher levels of bully victimization resulted in a 1.89 times increase in the odds of initiation in
Model 2. This effect was not seen in the other models and was only replicated in 1 out of 1000
bootstrap samples. Overall friendship and the number of close friends were not significantly
associated with initiation in late adolescence. The twin’s relationship with their caregiver was
also not a significant predictor of initiation.
Parental Psychosocial. Higher levels of caregiver perceived stress was associated with
increased odds of initiation in late adolescence for the adjusted models (M2=1.06, M3=1.09).
These results were not robust (effect replication < 2%). No prenatal support variables were
associated with initiation in late adolescence.
Gradient Boosted Modeling
Cross-validation. In the exploratory dataset, 10-fold cross validation was used to
identify the tree complexity (ie. interaction depth) for the GBM model. A graphical display of
the generalization error for each tree iteration is provided in Figure 2. An interaction depth of 6
(7 terminal nodes) minimized the cross-validated generalization error and was chosen as the
depth for the final model. A separate GBM was then conducted at an interaction depth of 6, with
cross-validation used to determine the appropriate number of decision trees to utilize (Figure 3).
ANTECEDENTS MARIJUANA INITIATION
20
Tree number 1,523 was found to minimize the generalization error and was used as the final
GBM model.
Relative influence of predictors. Figure 4 provides a graphical display of the relative
influence of the 61 variables used in the exploratory dataset GBM. The five most influential
variables for predicting if a participant will ever use marijuana were: Overall the friendship scale
(rel.inf=5.016), overall bully victimization scale (4.967), family socio-economic status (3.909),
caregiver perceived stress (3.594), and twin-report of positive parent-to-child affect (3.574). A
full listing of the relative influences of each variable are presented in supplementary Table S2.
Supplementary Figure S2 displays the relationships of each of these variables with the predicted
probability of ever use. Overall friendship, bully victimization, and caregiver perceived stress all
exhibit positive associations with the predicted probabilities of ever use indicating them as risk
factors, while socioeconomic status and parent-to-child positive affect are negatively associated
and considered protective against marijuana use. The least influential variables were variables
associated with substance use by the twin, peers, and mother during pregnancy.
Classification accuracy. The operating point that maximized the sensitivity and
specificity of the GBM predictions in the exploratory dataset was found to be 0.5259. Using this
value, participants with predicted probabilities of ever using marijuana > 0.5259 were classified
as having used marijuana. Using this cutpoint, the classification accuracy of the model was
assessed and found to be high in the exploratory dataset (82.6%) and moderate in the testing
dataset (65.2%). To determine if these models performed better than chance, Area Under the
Curve (AUC) was calculated using a bootstrapping technique described in the methods. Figure 5
displays the Receiver Operating Curve of these models and associated AUC’s. Both models
ANTECEDENTS MARIJUANA INITIATION
21
performed significantly better than change with the exploratory dataset having an AUC of 0.91
(95%CI 0.86, 0.95) and the testing dataset having an AUC of 0.71 (95%CI 0.64, 0.77).
Discussion
Impact of Attrition on Marijuana Outcomes
In this longitudinal study, the overall attrition rate was low across the ten year study
period such that the attrition rate of 22% equates to an approximately 6% loss to follow-up per
Wave. Despite this, attrition can have a meaningful impact on the outcome being studied,
particularly with regard to substance use, such that those using drugs are more likely to exit the
study early (Snow, Tebes, & Arthur, 1992; Hansen, Collins, Malotte, Johnson, & Fielding,
1985). In our study, we found that many of our predictors (~20%) were associated with attrition
and likely influence the prevalence estimates we obtain for marijuana useage. For example, peer
marijuana use was associated with attrition, making it likely that those who did not complete the
study may have done so due to their own marijuana use or factors associated with use. While
these variables were adjusted for in the fully adjusted models (assuming data were Missing At
Random), the impact is to underestimate the prevalence of marijuana initiation. Conversely, the
missing data inference strategy employed results in an overestimation of initiation by the ability
to infer initiation from a response at any Wave endorsing marijuana use. To state that a
participant never initiated requires being able to know their Wave 5 responses and thus
underestimates this group due to attrition. Differences between the two outcomes (initiation age
group and ever use) in the proportions of never users/initiates is primarily due to differences in
these attrition patterns. Overall our prevalence of marijuana initiation (56.5%) was slightly
higher than the national average lifetime prevalence of 50% (NSDUH, 2012; Shi, 2013). Lastly,
while this study examines attrition with respect to the outcome, the predictors themselves may
ANTECEDENTS MARIJUANA INITIATION
22
have meaningful attrition patterns with respect to each other that affect the overall reliability of
the models. While examining such attrition patterns is beyond the scope of this paper, more work
needs to be done to understand how attrition in longitudinal studies may obscure the outcomes
we study.
Exploratory Analysis and Methods
There has been much focus in the literature on the importance of replication and the
reporting of results as we acknowledge that many published results are either of extreme
magnitudes not likely to be found in the population (Gelman & Weakliem, 2009) or a matter of
false positives (Gelman & Loken, 2013). While recent efforts to combat the reproducibility crisis
have focused on “The New Statistics” as proposed by Cumming (2013), this is simply a
repackaging of Null Hypothesis Significance Testing which is prone to the same problems of
reporting false positives and unlikely magnitudes as statistically significant results. This study
takes the perspective that much the analyses we conduct are exploratory, such that we rarely
have hypotheses specific enough to proclaim an expected magnitude. Similarly we do not limit
our examinations of the data solely to our a priori beliefs. This study makes use of a twin sample
to conduct an exploratory analysis in a principled manner. Rather than conducting the analysis
with an arbitrary correction for multiple comparisons, the MLoR results provide estimates of the
cross-sample reproducibility of the effects as a means to discern a true positive from a false
positive. The estimate of cross-sample reproducibility presented in this paper was called “effect
replication”, however what makes for a convincing magnitude of effect replication is not clear.
In one sense, effect replication can be viewed as a type of power estimation that requires the
observed effect in the cross-sample to be not only statistically significant but also of similar
magnitude to the original sample. In this sense, perhaps standard power values are applicable
ANTECEDENTS MARIJUANA INITIATION
23
(eg. 80%) though given that detected effects are likely to be of larger magnitude than in the
population, a higher power threshold would be more desirable. What is clear is that effects that
cannot be replicated within the study sample are unlikely to be reproduced in an entirely new
sample. An indication of this would be effect replication values less than 50%.
An additional concern of this study is accuracy in classification. While classical statistical
modeling approaches are concerned with parsimony, often this simplicity comes at a cost of
accuracy. To this end, the authors acknowledge that behavior is non-linear and complex with
numerous variables interacting with one another in high dimensions. As such, the more accurate
models of behavior will be those that incorporate these elements. Decision Tree Analysis is a
technique that has been around since the late 1950’s (Belson, 1959; McArdle, 2013) that is
especially adept at modeling non-linearity and high order interactions. The gradient boosted
modeling conducted in this study is one of many methods in statistical (machine) learning based
on decision trees that has recently been re-discovered as part of the “data science” and “big data”
movements. These atheoretical algorithm based techniques are designed to produce the best
predictions given the data. As such, they can have great utility in developing highly accurate
models of behavior. As these techniques are beginning to be applied with greater frequency in
the scientific research world, it is important to develop a framework for the use of these
techniques in a principled way.
An important contribution of this work to this framework is the use of an operating point
to determine the predicted probability for binary classification (see Methods section). Often
binary classification is determined based upon a predicted probability of greater or less than 0.5,
however for many classification models this value may not maximize the classification accuracy.
Indeed, some prediction models may not even have predicted probabilities that exist on one side
ANTECEDENTS MARIJUANA INITIATION
24
of the 0.5 demarcation. Instead, by finding the operating point which maximizes model
sensitivity and specificity, one is able to maximize classification accuracy. Alternately, the
equation for the operating point can be modified to weight false positives and negatives
unequally to create classifications that favor sensitivity over specificity or vice versa (ie. altering
the cost of misclassification).
An additional contribution of this study was the use of a cross-sample (testing dataset) to
determine how well the model developed in the exploratory dataset would generalize to a
random sample from the same study. This is an especially important aspect of utilizing these
statistical learning techniques, as high levels of classification accuracy should be expected in the
data the model is being developed on. Given such, high accuracy in the exploratory data is not
the end goal, but rather determining how generalizable the model would be to novel data.
Predictors of Initiation Age Group.
Most variables that were significant in the MLoR had low effect replications (<50%).
These variables will not be mentioned in the discussion as they are considered to be spurious
associations that have a low chance of replicating in a novel sample.
Monozygotic twins were found to be at greater risk on marijuana initiation in early
adolescence. Studies demonstrate that the social environment of monozygotic twins differs from
that of dizygotic’s, typically such that the monozygotics have a greater degree of shared
experiences and peer groups (Felson, 2014). Given this, it seems likely that this increase risk is
resultant from groups differences on another (perhaps unmeasured) variable that confers risk. It
is likely then that these differences are mediated by this third unexplained variable which appears
in greater magnitude for one of the Zygosity groups.
ANTECEDENTS MARIJUANA INITIATION
25
Twin alcohol use (having a sip) was predictive of early initiation, which is salient with
the literature suggesting that early initiation of tobacco or alcohol use is a risk factor for later
marijuana use (Kosterman et al., 2000; Ellickson et al., 2004). This result was only robust in the
unadjusted models suggesting that in the cross-sample socio-demographic and other covaried out
factors mediate these observed differences. Given that there is commonality in many of the risk
factors for these substances (Hawkins, Catalano, & Miller, 1992), this result and it’s lack of
robustness after covariate adjustment is not surprising.
Prenatal psychosocial support has been linked to positive infant health outcomes such as
greater birth weight and fetal growth (Feldman, Dunkel-Schetter, Sandman, & Wadhwa, 2000),
however no study to the authors’ knowledge has examined the impact of prenatal psychosocial
support on substance use in the offspring. Our results indicate that having high levels of support
from the family as well as having planned the pregnancy are protective factors against marijuana
initiation in early and mid-adolescence. These results are most likely capturing the effects of
prenatal stress, which have been linked to emotional and behavioral problems in the offspring
(O’Connor, Heron, Golding, & Glover, 2003). Additionally, it is likely that there are genetic
factors at play in determining the mother’s predisposition to viewing themselves as “supported”.
These genetic factors inherited by the offspring could play a large role in determining substance
use (Kendler & Prescott, 2006).
The overall friendship scale from the Childhood Friendship Questionnaire was a robust
predictor of marijuana initiation in the MLoR. Results indicated that reporting higher levels of
social interaction was related to increased risk of marijuana initiation. The literature on substance
use in college students seems to support this notion with studies suggesting that Greek life
involvement or simply greater social networks grant greater exposure to and opportunity for
ANTECEDENTS MARIJUANA INITIATION
26
marijuana use (Suerken et al., 2014; Bell et al., 1997). Alternately, these results could reflect that
social competency is required to facilitate a drug transaction, and thus individuals with these
skills are at greater risk for initiation. Recent research on developmental trajectories of kids
viewed as “popular” suggest poorer long term outcomes, particularly with regard to drugs and
alcohol use (Allen, Schad, Oudekerk, & Chango, 2014). Adolescent popularity is typically
associated with minor-delinquency and romantic involvement, both of which seek to distinguish
the adolescent as mature. While socially normative, it is understood that the expression of these
behaviors may be used to mask personal deficits in confidence and maturity, particularly when
expressed prematurely in development (Galambos & Tilton-Weaver, 2000). As such, high
levels of “friendship” here may be indicative of the pseudomature behavior described by Allen et
al. and thus reflecting a vulnerability to marijuana initiation for these participants.
Bully victimization has been shown to be a predictor of substance use in adolescence
(Tharp-Taylor, Haviland, & D’Amico, 2009; Kaltiala-Heino, Rimpelä, M., Marttunen, Rimpelä,
A., & Rantanen, 1999). In our study, bully victimization increased the risk for marijuana
initiation in early adolescence, though this result was only robust in the unadjusted model. This
increased risk of initiation may be due to the use of marijuana as an emotional coping strategy
(Comeau, Stewart, & Loba, 2001), though evidence of self-medication as a motivating factor for
early initiates is lacking (Chassin et al, 1999). Alternately this result may reflect a common
correlate between marijuana initiation and bully victimization such that differences in behavior
or personality that make it more likely to be a victim of bullying also make it more likely to use
marijuana (Kaltiala-Heino, Rimpelä, M., Rantanen, Rimpelä, A., & Rantanen, 1999; Luk, Wang,
& Simons-Morton, 2010).
ANTECEDENTS MARIJUANA INITIATION
27
Predictors of Ever Use
Gradient boosting was used to identify variables that were influential in predicting if a
participant would ever use marijuana.
The model presented in this paper excludes ethno-racial group as potential predictor (thus
only 61 instead of 62 predictors). This is because when ethno-racial group is used in the GBM,
this variable is one of the strongest predictors of initiation, but only because 1 Asian participant
had ever used marijuana. There is little utility in a model that predicts marijuana use simply
because an individual is not of Asian descent, particularly when the sample size for Asians is
very small in the dataset (N =16). While the sample size is too small to make inferences on in our
sample, many studies find that cultural differences can play a large role in the prevalence of
initiation. In China, for example, the lifetime prevalence of marijuana use is less than 1% (Shi,
2013). Studies looking at ethno-racial difference in marijuana initiation in the United States
routinely find that Asian participants have lower rates of initiation (Kosterman et al, 2000;
Jessor, 1976; So & Wong, 2006).
The top predictors for marijuana use were some of the same variables identified in the
MLoR analyses, such that the overall friendship scale and bully victimization scale were the
most influential variables in predicting initiation. The robustness of the friendship scale in the
MLoR and its identification as the most influential predictor in the GBM are taken to be
indications of the importance of this variable on initiation. A description of the mechanisms by
which the friendship and bully victimization scales may be operating on marijuana initiation are
described in the discussion of the initiation age group results. It is interesting to note that for
bully victimization (scale range 1-5), even a small amount of bullying resulted in a greater than
ANTECEDENTS MARIJUANA INITIATION
28
20% increase in the probabilities of initiation. In the model, values greater than 1.5 on the bully
victimization scale would result in classification as someone who has used marijuana.
Socioeconomic status (SES) was found to be a highly influential variable in predicting
initiation. While previous studies have identified the influential role of this variable, many
suggest increased risk at either extreme (affluence and poverty) of socioeconomic level (Crabbe,
2002; Suerken et al., 2014). Our results support the notion that high SES is a protective factor
against marijuana initiation, while low SES conferred a very slight risk of initiation. These
results are in contrast with previous findings regarding affluence and marijuana use (Suerken et
al., 2014), though this difference may be due to differences in SES assessment. For example, the
current study assesses SES for the family at ages 9-10, while Suerken et al. examined the income
and spending money available to adolescents as predictors of use.
Caregiver perceived stress was found to increase the probability of having used marijuana
and was the fourth most influential variable from the GBM. Numerous studies document that
parenting style and family management are strong predictors of marijuana initiation (Kosterman
et al., 2000; Siegel et al., 2013; Hawkins et al., 1992; Wills &Cleary, 1996). The perceived stress
scale not only captures recent sources of stress (eg. “In the last month, how often were you
behind with things you needed to do?”) but also a predisposition of the caregiver (eg. “How
often have you felt things were going your way?”). It is likely the later that this result is
capturing which would also be confounded by the genetic contributions of personality and
temperament. Indeed it should not be surprising that the offspring of caregivers who report
higher levels of stress and inability to cope with stressful events would also exhibit similar
deficits either as learned behavior or through genetic mechanisms that would lead to marijuana
initiation as a means to cope with perceived stress.
ANTECEDENTS MARIJUANA INITIATION
29
Greater positive parent-to-child affect, reported by the twin, was protective against
initiation and was the fifth most influential variable in the GBM. This finding concurs with
previous research that suggests low bonding as a risk factor and supportive parenting as a
protective factor for marijuana initiation (Kosterman et al., 2000; Siegel et al., 2013). While
lower levels of positive affect were associated with a moderate risk of initiation, high levels of
positive affect resulted in a ~15 point drop in the predicted probabilities of initiation , indicating
positive affect as a major protective factor against initiation.
The least influential variables in the GBM were related to twin and peer substance use as
well as mother’s substance use during pregnancy. These findings are in contrast with studies that
have found peer use to be a significant risk factor of substance use (Fowler et al. 2007;Creemers
et al., 2010; Crabbe, 2002). Similarly, while previous studies have suggested prenatal exposure
to alcohol increases pharmacological vulnerability and thus propensity for abuse (Sher et al.,
2005), our study found little association between prenatal substance exposure and marijuana
initiation. This discrepancy in results may have to do with low variability in these predictors,
particularly with regard to twin and peer use. Given that this study assessed these variables at
ages 9-10, very few twins were endorsing substance use items. Future analyses looking at peer
group influences in early adolescence may help clarify the veracity of this finding, particularly
with regard to peer marijuana use.
Model accuracy. This study used variables from late childhood (9-10 years) to predict a
relatively diffuse outcome such as having ever used marijuana in adolescence. The model
developed was accurate (82.6% and 65.2% classification accuracy) and statistically significant in
both datasets. While the technique employed is designed to produce accurate classification, there
is little doubt that if the outcome were more specific (eg. regular marijuana use or diagnosed
ANTECEDENTS MARIJUANA INITIATION
30
substance use disorder) an even more accurate model could be developed as there are likely
many benign associations captured within this model.
Strengths and Limitations
This study has numerous strengths such that 1) all predictors precede marijuana initiation
temporally 2) a split-sample is used to evaluate the robustness of the findings and 3) a classical
(MLoR) and modern (GBM) approach to identifying predictors of initiation are used.
A potential weakness of this study is in the assessment of initiation age groups. Because
subjects must have completed Wave 5 or have data that would allow for inferring the Wave 5
marijuana use variable, some subjects are excluded from the analysis due to attrition. While an
attempt was made to adjust for variables associated with attrition, a more powerful analysis
would be to directly model attrition by incorporating censoring. Similarly, using discrete age
groups for initiation may not be as sensitive as using age at initiation. Future analyses will
examine initiation in the context of survival analysis to account for censoring and a more precise
age at initiation.
Future Directions
While this paper identified the predictors for marijuana initiation, an effort must be made
to differentiate predictors of experimentation which may be a part of normative adolescent
development and predictors of initiation leading to pathological use. Future work will focus on
making this distinction. Additional work needs to be done on understanding how individuals who
initiate marijuana in early adolescence differ from those who initiate later in life. It is yet
uncertain if Cloninger’s alcoholic subtype models will apply to marijuana users. As the study
participants of the RFAB are moving out of adolescence and into adulthood, we look forward to
answering these questions.
ANTECEDENTS MARIJUANA INITIATION
31
Acknowledgements
This work was supported by a grant from the National Institute of Mental Health (NIMH)
grant number R01-MH58354. The authors would like to acknowledge M. Dudzinsky, K. Gomez,
& G. Quick for their contributions to the study management and the data collection. N. Jackson
thanks A.Howerter and M. Grandner for providing mentorship at different phases of his career
and H. Wang for stimulating conversations regarding Receiver Operating Curves.
ANTECEDENTS MARIJUANA INITIATION
32
References
Allen, J. P., Schad, M. M., Oudekerk, B., & Chango, J. (2014). What Ever Happened to the “Cool”
Kids? Long‐Term Sequelae of Early Adolescent Pseudomature Behavior. Child
Development.
Arseneault, L., Cannon, M., Witton, J., & Murray, R. M. (2004). Causal association between
cannabis and psychosis: examination of the evidence. The British Journal of
Psychiatry, 184(2), 110-117.
Baker, L. A., Barton, M., & Raine, A. (2002). The southern California twin register at the University
of Southern California. Twin Research, 5(05), 456-459.
Bell, R., Wechsler, H., & Johnston, L. D. (1997). Correlates of college student marijuana use:
Results of a US national survey. Addiction, 92(5), 571-581.
Belson, W. A. (1959). Matching and prediction on the principle of biological classification. Applied
statistics, 65-75.
Bezdjian, S., Raine, A., Baker, L. A., & Lynam, D. R. (2011). Psychopathic personality in children:
genetic and environmental contributions. Psychological medicine, 41(03), 589-600.
Brook, J. S., Adams, R. E., Balka, E. B., & Johnson, E. (2002). Early adolescent marijuana use:
Risks for the transition to young adulthood. Psychological Medicine, 32(1), 79-92.
Brook, J. S., Balka, E. B., & Whiteman, M. (1999). The risks for late adolescence of early adolescent
marijuana use. American Journal of Public Health, 89(10), 1549-1554.
Brook, J. S., Stimmel, M. A., Zhang, C., & Brook, D. W. (2008). The association between earlier
marijuana use and subsequent academic achievement and health problems: A longitudinal
study. The American Journal on Addictions,17(2), 155-160.
ANTECEDENTS MARIJUANA INITIATION
33
Chassin, L., Pitts, S. C., DeLucia, C., & Todd, M. (1999). A longitudinal study of children of
alcoholics: predicting young adult substance use disorders, anxiety, and depression. Journal
of Abnormal Psychology, 108(1), 106.
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. Social Psychiatry and Psychiatric Epidemiology,37(5), 199-206.
Colorado Constitutional Amendment LXIV (enacted 2014).
Comeau, N., Stewart, S. H., & Loba, P. (2001). The relations of trait anxiety, anxiety sensitivity, and
sensation seeking to adolescents' motivations for alcohol, cigarette, and marijuana
use. Addictive behaviors, 26(6), 803-825.
Crabbe, J. C. (2002). Genetic contributions to addiction*. Annual review of psychology, 53(1), 435-
462.
Creemers, H. E., Dijkstra, J. K., Vollebergh, W. A., Ormel, J., Verhulst, F. C., & Huizink, A. C.
(2010). Predicting life‐time and regular cannabis use during adolescence; the roles of
temperament and peer substance use: the TRAILS study. Addiction, 105(4), 699-708.
Cumming, G. (2013). Understanding the new statistics: Effect sizes, confidence intervals, and meta-
analysis. Routledge.
DeWit, D. J., Adlaf, E. M., Offord, D. R., & Ogborne, A. C. (2000). Age at first alcohol use: a risk
factor for the development of alcohol disorders. American Journal of Psychiatry, 157(5),
745-750.
DeSimone, J. (2002). Illegal drug use and employment. Journal of Labor Economics, 20(4), 952-
977.
ANTECEDENTS MARIJUANA INITIATION
34
Ellickson, P. L., Tucker, J. S., Klein, D. J., & Saner, H. (2004). Antecedents and outcomes of
marijuana use initiation during adolescence. Preventive medicine,39(5), 976-984.
Feldman, P. J., Dunkel-Schetter, C., Sandman, C. A., & Wadhwa, P. D. (2000). Maternal social
support predicts birth weight and fetal growth in human pregnancy. Psychosomatic
Medicine, 62(5), 715-725.
Felson, J. (2014). What can we learn from twin studies? A comprehensive evaluation of the equal
environments assumption. Social science research, 43, 184-199.
Fergusson, D. M., & Boden, J. M. (2008). Cannabis use and later life outcomes.Addiction, 103(6),
969-976.
Finnell, W. S., & Jones, J. D. (1975). Marijuana, alcohol and academic performance. Journal of
Drug Education, 5(1), 13-21.
Flory, K., Lynam, D., Milich, R., Leukefeld, C., & Clayton, R. (2004). Early adolescent through
young adult alcohol and marijuana use trajectories: Early predictors, young adult outcomes,
and predictive utility. Development and psychopathology, 16(01), 193-213.
Fowler, T., Shelton, K., Lifford, K., Rice, F., McBride, A., Nikolov, I., ... & Van Den Bree, M.
(2007). Genetic and environmental influences on the relationship between peer alcohol use
and own alcohol use in adolescents. Addiction,102(6), 894-903.
Friedman, J. H. (2001). Greedy function approximation: a gradient boosting machine. Annals of
Statistics, 1189-1232.
Friedman, J. H. (2002). Stochastic gradient boosting. Computational Statistics & Data
Analysis, 38(4), 367-378.
ANTECEDENTS MARIJUANA INITIATION
35
Galambos, N. L., & Tilton-Weaver, L. C. (2000). Adolescents' psychosocial maturity, problem
behavior, and subjective age: In search of the adultoid.Applied Developmental Science, 4(4),
178-192.
Gelman, A., & Loken, E. (2013). The garden of forking paths: Why multiple comparisons can be a
problem, even when there is no “fishing expedition” or “p-hacking” and the research
hypothesis was posited ahead of time. Downloaded January, 30, 2014.
Gelman, A., & Weakliem, D. (2009). Of beauty, sex and power. American Scientist, 97(4), 310-316.
Hansen, W. B., Collins, L. M., Malotte, C. K., Johnson, C. A., & Fielding, J. E. (1985). Attrition in
prevention research. Journal of Behavioral Medicine, 8(3), 261-275.
Hare, R. D. (1991). The Hare psychopathy checklist-revised . Toronto, Ontario: Multi-Health
Systems.
Hawkins, J. D., Catalano, R. F., & Miller, J. Y. (1992). Risk and protective factors for alcohol and
other drug problems in adolescence and early adulthood: implications for substance abuse
prevention. Psychological bulletin, 112(1), 64.
Hosmer, D. W., & Lemeshow, S. (1989). Applied regression analysis. New York, John Willey.
Jessor, R. (1976). Predicting time of onset of marijuana use: A developmental study of high school
youth. Journal of Consulting and Clinical Psychology, 44(1), 125.
Kaltiala-Heino, R., Rimpelä, M., Marttunen, M., Rimpelä, A., & Rantanen, P. (1999). Bullying,
depression, and suicidal ideation in Finnish adolescents: school survey. Bmj, 319(7206), 348-
351.
Kaltiala-Heino, R., Rimpelä, M., Rantanen, P., & Rimpelä, A. (2000). Bullying at school—an
indicator of adolescents at risk for mental disorders. Journal of adolescence, 23(6), 661-674.
ANTECEDENTS MARIJUANA INITIATION
36
Kendler, K. S., & Prescott, C. A. (2006). Genes, environment, and psychopathology. New York:
Guilford.
Kokkevi, A., Richardson, C., Florescu, S., Kuzman, M., & Stergar, E. (2007). Psychosocial
correlates of substance use in adolescence: A cross-national study in six European
countries. Drug and alcohol dependence, 86(1), 67-74.
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. American
Journal of Public Health, 90(3), 360.
Lisdahl, K. M., Wright, N. E., Medina-Kirchner, C., Maple, K. E., & Shollenbarger, S. (2014).
Considering Cannabis: The Effects of Regular Cannabis Use on Neurocognition in
Adolescents and Young Adults. Current Addiction Reports,1(2), 144-156.
Luby, J. L., Svrakic, D. M., McCallum, K., Przybeck, T. R., & Cloninger, C. R. (1999). The Junior
Temperament and Character Inventory: preliminary validation of a child self-report
measure. Psychological reports, 84(3c), 1127-1138.
Luk, J. W., Wang, J., & Simons-Morton, B. G. (2010). Bullying victimization and substance use
among US adolescents: mediation by depression. Prevention Science, 11(4), 355-359.
Lynam, D. R. (1997). Pursuing the psychopath: capturing the fledgling psychopath in a nomological
net. Journal of abnormal psychology, 106(3), 425.
Lynskey, M., & Hall, W. (2000). The effects of adolescent cannabis use on educational attainment: a
review. Addiction, 95(11), 1621-1630.
Macleod, J., Oakes, R., Copello, A., Crome, I., Egger, M., Hickman, M., ... & Smith, G. D. (2004).
Psychological and social sequelae of cannabis and other illicit drug use by young people: a
ANTECEDENTS MARIJUANA INITIATION
37
systematic review of longitudinal, general population studies. The Lancet, 363(9421), 1579-
1588.
McArdle, J.J. (2013). Exploratory Data Mining Using Decision Trees. In McArdle, J. J., &
Ritschard, G. (Eds.), Contemporary Issues in Exploratory Data Mining in the Behavioral
Sciences (pp. 3-47). New York, NY: Routledge.
Meier, M. H., Caspi, A., Ambler, A., Harrington, H., Houts, R., Keefe, R. S, & Moffitt, T. E.
(2012). Persistent cannabis users show neuropsychological decline from childhood to
midlife. Proceedings of the National Academy of Sciences,109(40), E2657-E2664.
Moore, T. H., Zammit, S., Lingford-Hughes, A., Barnes, T. R., Jones, P. B., Burke, M., & Lewis, G.
(2007). Cannabis use and risk of psychotic or affective mental health outcomes: a systematic
review. Lancet, 370, 319-28.
NSDUH (2012).National Survey on Drug Use and Health, 2011. ICPSR34481-v1. In (Vol.
2013).Ann Arbor, MI: Inter-university Consortium for Political and Social Research: United
States Department of Health and Human Services. Substance Abuse and Mental Health
Services Administration (Center for Behavioral Health Statistics and Quality)
O'Connor, T. G., Heron, J., Golding, J., & Glover, V. (2003). Maternal antenatal anxiety and
behavioural/emotional problems in children: a test of a programming hypothesis. Journal of
Child Psychology and Psychiatry, 44(7), 1025-1036.
Oldfield, R. C. (1971). The assessment and analysis of handedness: the Edinburgh
inventory. Neuropsychologia, 9(1), 97-113.
Pacula, R. L., & Kilmer, B. (2003). Marijuana and crime: Is there a connection beyond
prohibition? (No. w10046). National Bureau of Economic Research.
ANTECEDENTS MARIJUANA INITIATION
38
Pedersen, W., & Skardhamar, T. (2010). Cannabis and crime: Findings from a longitudinal
study. Addiction, 105(1), 109-118.
Popovici, I., French, M. T., Pacula, R. L., Maclean, J. C., & Antonaccio, O. (2014). Cannabis Use
and Antisocial Behavior among Youth. Sociological Inquiry, 84(1), 131-162.
Remnick, D. (2014, January 27). Going the Distance. The New Yorker. Retrieved from
http://www.newyorker.com/reporting/2014/01/27/140127fa_fact_remnick
Ridgeway, G. (2013). gbm: Generalized Boosted Regression Models. R package version 2.0-8.
http://CRAN.R-project.org/package=gbm.
Ringel, J. S., Ellickson, P. L., & Collins, R. L. (2006). The relationship between high school
marijuana use and annual earnings among young adult males.Contemporary Economic
Policy, 24(1), 52-63.
SAMHSA (2012).Results from the 2011 National Survey on Drug Use and Health: Summary of
National Findings, NSDUH Series H-44, HHS Publication No. (SMA) 12-4713. Rockville,
MD: Substance Abuse and Mental Health Services Administration.
Semple, D. M., McIntosh, A. M., & Lawrie, S. M. (2005). Cannabis as a risk factor for psychosis:
systematic review. Journal of psychopharmacology, 19(2), 187-194.
Sher, K. J., Grekin, E. R., & Williams, N. A. (2005). The development of alcohol use
disorders. Annu. Rev. Clin. Psychol., 1, 493-523.
Shi, Y. (2014). At high risk and want to quit: Marijuana use among adults with depression or serious
psychological distress. Addictive behaviors, 39(4), 761-767.
Siegel, J. T., Crano, W. D., Alvaro, E. M., Lac, A., Hackett, J. D., & Hohman, Z. P. (2014).
Differentiating Common Predictors and Outcomes of Marijuana Initiation: A Retrospective
Longitudinal Analysis. Substance use & misuse, 49(1-2), 30-40.
ANTECEDENTS MARIJUANA INITIATION
39
Snow, D. L., Tebes, J. K., & Arthur, M. W. (1992). Panel attrition and external validity in adolescent
substance use research. Journal of consulting and clinical psychology, 60(5), 804.
So, D. W., & Wong, F. Y. (2006). Alcohol, drugs, and substance use among Asian-American college
students. Journal of psychoactive drugs, 38(1), 35-42.
Suerken, C. K., Reboussin, B. A., Sutfin, E. L., Wagoner, K. G., Spangler, J., & Wolfson, M. (2014).
Prevalence of marijuana use at college entry and risk factors for initiation during freshman
year. Addictive behaviors, 39(1), 302-307.
Tharp-Taylor, S., Haviland, A., & D'Amico, E. J. (2009). Victimization from mental and physical
bullying and substance use in early adolescence. Addictive behaviors, 34(6), 561-567.
US Census Bureau (2002). Los Angeles County Census 2000 Statistical Compilation. Washington,
DC: US Census Bureau
Wills, T. A., & Cleary, S. D. (1996). How are social support effects mediated? A test with parental
support and adolescent substance use. Journal of personality and social psychology, 71(5),
937.
ANTECEDENTS MARIJUANA INITIATION
40
Tables
Table 1: Distributions of Marijuana Outcomes
Exploratory Data
N=351
Testing Data
N=335
Initiation Age
Groups
Never Initiated 54.7% 52.5%
Early Adolescence 17.1% 16.4%
Mid- Adolescence 20.5% 21.5%
Late Adolescence 7.7% 9.6%
Exploratory Data
N=432
Testing Data
N=414
Ever Use
Never Used 44.4% 42.5%
Have Used 56.6% 57.5%
ANTECEDENTS MARIJUANA INITIATION
41
Table 2: Bivariate Associations with Marijuana Initiation Age Groups
Variable Type Variable Name Values
Never
Initiated
N=192
Early
Adolescen
ce
N=60
Mid-
Adolescen
ce
N=72
Late
Adolescen
ce
N=27
P
Socio-
Demographics
Age 9.55 ± 0.57 9.75 ± 0.65 9.71 ± 0.61 9.65 ± 0.68 §0.193
Gender
1:M 40.10% 46.67% 55.56% 51.85%
§0.129
2:F 59.90% 53.33% 44.44% 48.15%
Race
1: Cau 28.13% 28.81% 32.39% 44.44%
§0.042
2: His 30.21% 49.15% 39.44% 25.93%
3: AA 13.02% 10.17% 9.86% 14.81%
4: Asian 7.81% 1.69% 0.00% 0.00%
7: Multi 20.83% 10.17% 18.31% 14.81%
Zygosity
0: MZ 44.79% 23.33% 41.67% 22.22%
§0.006
1: DZ 55.21% 76.67% 58.33% 77.78%
Socioeconomic Status 43.6 ± 12.7 39.7 ± 11.2 44.2 ± 12.3 43.7 ± 13.6 0.254
Neighborhood Characteristics Score 1.58 ± 0.61 1.77 ± 0.73 1.61 ± 0.52 1.45 ± 0.44 §0.173
Edinburg Hand Score 67.4 ± 45.0 68.9 ± 46.7 58.5 ± 55.4 57.7 ± 51.9 0.546
Handedness %Right 94.15% 93.33% 88.89% 92.59% 0.535
Twin
Substance Use
Puff of Cig %Yes 0.00% 4.35% 1.79% 0.00% 0.111
Sip of Alcohol %Yes 16.03% 32.61% 25.00% 26.09% §0.099
More than Sip %Yes 1.53% 6.52% 1.79% 0.00% 0.211
Whole Drink %Yes 0.00% 0.00% 0.00% 0.00%
Peer
Substance Use
PA: Have Smoked
Cigs
%Yes 0.76% 9.09% 3.57% 0.00% 0.026
PA: Have Drunk Alc %Yes 1.53% 4.76% 12.50% 4.55% §0.016
PA: Have used Mar %Yes 0.00% 2.27% 1.79% 4.55% 0.233
PA: Have used Drugs %Yes 0.00% 0.00% 1.79% 0.00% 0.317
PDI: Smoked Cigs 1.00 ± 0.00 1.31 ± 0.54 1.22 ± 0.56 1.14 ± 0.36 <.001
PDI: Sip of Alcohol 1.17 ± 0.50 1.54 ± 0.74 1.46 ± 0.78 1.00 ± 0.00 0.001
PDI: More than Sip 1.40 ± 0.52 1.46 ± 0.88 1.57 ± 0.94 NA 0.893
PDI: Whole Drink 1.40 ± 0.55 1.50 ± 1.00 1.86 ± 0.90 NA 0.619
PDI: Marijuana Use 1.00 ± 0.00 1.13 ± 0.34 1.14 ± 0.41 1.05 ± 0.22 0.007
PDI: Drugs 1.00 ± 0.00 1.11 ± 0.31 1.12 ± 0.50 1.00 ± 0.00 0.046
Mom
Substance Use
in Pregnancy
MHQ: Smoke Cigs %Yes 5.43% 11.36% 7.41% 4.35% 0.556
MHQ: Consume
Alcohol
%Yes 8.53% 9.09% 12.73% 17.39% 0.552
MHQ: Drug Use %Yes 1.56% 6.67% 5.45% 0.00% 0.209
Prenatal
Support
Pregnancy Planned
1: No 36.51% 61.90% 49.06% 45.45%
§0.081 2: Some 6.35% 9.52% 7.55% 4.55%
3: Yes 57.14% 28.57% 43.40% 50.00%
How depressed when found out
preg?
1.67 ± 1.21 1.74 ± 1.23 1.89 ± 1.28 1.59 ± 0.96 0.695
How much did you want this preg? 4.05 ± 1.23 3.68 ± 1.40 3.96 ± 1.26 4.05 ± 1.43 0.463
How happy was your partner? 4.14 ± 1.04 3.78 ± 1.11 3.87 ± 1.19 4.10 ± 0.89 §0.185
How supportive was your family? 4.36 ± 0.86 3.69 ± 1.12 4.06 ± 1.01 4.14 ± 1.13 §0.001
Depression after birth? 1.85 ± 1.18 1.88 ± 1.17 2.00 ± 1.14 1.73 ± 0.94 0.789
Marital
Satisfaction
Amount of Time Spent Together 2.21 ± 0.96 2.60 ± 1.35 2.44 ± 1.05 2.45 ± 0.96 0.238
Amount of effort partner puts into
relationship
2.10 ± 1.07 2.37 ± 1.07 2.23 ± 1.07 2.68 ± 1.09 0.115
Amount of attention partner gives
you
2.03 ± 1.06 2.13 ± 1.07 2.21 ± 1.06 2.45 ± 0.74 0.323
Satisfied sex life/intimacy 2.05 ± 1.07 1.93 ± 1.11 2.14 ± 1.01 2.48 ± 0.98 0.300
Overall, how happy are you in your 1.82 ± 0.94 1.93 ± 1.11 1.74 ± 0.89 2.45 ± 0.80 §0.024
ANTECEDENTS MARIJUANA INITIATION
42
relationship
Childhood
Psychopathy
Scale
Hare Factor 1:
(shallow affect, superficial charm,
manipulativeness, lack of empathy)
0.207 ± 0.120 0.225 ± 0.107 0.197 ± 0.079 0.233 ± 0.157 0.467
Hare Factor 2:
(criminal versatility, impulsiveness,
irresponsibility, poor behavior
controls, juvenile delinquency)
0.214 ± 0.143 0.240 ± 0.147 0.236 ± 0.138 0.212 ± 0.145 0.616
Bezdjian Factor 1: (callousness-
disinhibited)
0.246 ± 0.137 0.267 ± 0.129 0.266 ± 0.133 0.237 ± 0.149 0.659
Bezdjian Factor 2: (manipulative-
decietful)
0.167 ± 0.133 0.191 ± 0.125 0.156 ± 0.083 0.207 ± 0.158 0.266
Junior
Temperament
(T) and
Character (C)
Inventory
Novelty Seeking (T) 0.226 ± 0.134 0.251 ± 0.157 0.230 ± 0.141 0.213 ± 0.141 0.708
Harm Avoidance (T) 0.481 ± 0.189 0.507 ± 0.189 0.474 ± 0.219 0.457 ± 0.192 0.745
Reward Depend (T) 0.526 ± 0.196 0.514 ± 0.180 0.519 ± 0.176 0.575 ± 0.216 0.629
Persistence (T) 0.693 ± 0.220 0.680 ± 0.198 0.687 ± 0.236 0.775 ± 0.185 0.330
Self-Directed(C) 0.712 ± 0.159 0.680 ± 0.166 0.689 ± 0.169 0.720 ± 0.149 0.576
Cooperative (C) 0.812 ± 0.136 0.784 ± 0.136 0.810 ± 0.135 0.847 ± 0.108 0.317
Parent to Child
Affect
Parent Report-Positive 4.18 ± 0.41 4.15 ± 0.41 4.05 ± 0.42 4.14 ± 0.43 0.316
Parent Report-Negative 2.21 ± 0.56 2.34 ± 0.72 2.44 ± 0.58 2.43 ± 0.61 §0.063
Twin Report-Positive 3.95 ± 0.60 3.96 ± 0.58 3.82 ± 0.60 4.07 ± 0.50 0.339
Twin Report-Negative 2.10 ± 0.76 2.30 ± 0.85 2.13 ± 0.74 2.04 ± 0.71 0.460
Child
Friendship
Questionnaire
Friendship Initiator Scale 3.59 ± 1.05 4.09 ± 1.01 3.94 ± 1.08 3.86 ± 1.00 0.026
Friendship Receiver Scale 3.35 ± 1.07 4.00 ± 0.88 3.65 ± 0.95 3.75 ± 0.91 0.002
Overall Friendship Scale 3.44 ± 0.95 4.01 ± 0.80 3.74 ± 0.85 3.79 ± 0.89 §0.002
Bullied @ School: Indirect (exile)
Scale
1.80 ± 0.93 1.97 ± 1.20 2.36 ± 1.34 2.25 ± 1.13 0.011
Bullied @ School: Verbal Scale 1.66 ± 0.82 2.18 ± 1.06 1.91 ± 1.05 1.98 ± 1.22 0.013
Bullied @ School: Physical Scale 1.35 ± 0.62 1.54 ± 0.75 1.55 ± 1.04 1.75 ± 1.15 0.101
Bullied @ School: Overall Scale 1.56 ± 0.66 1.96 ± 0.82 1.85 ± 0.89 1.93 ± 1.04 0.008
Bullied in Neighborhood Scale 1.48 ± 0.73 1.76 ± 0.96 1.56 ± 0.93 1.72 ± 1.05 0.229
Overall Bullied Scale 1.54 ± 0.64 1.89 ± 0.80 1.77 ± 0.85 1.85 ± 0.98 §0.023
Parental Disapproval of Friends
Scale
1.35 ± 0.50 1.55 ± 0.60 1.41 ± 0.43 1.43 ± 0.85 0.209
# Of Close Friends 5.22 ± 2.81 6.13 ± 2.72 5.91 ± 2.88 6.48 ± 2.73 §0.084
# Of Casual Friends 6.66 ± 9.82 11.41 ± 18.26 9.00 ± 16.77 8.96 ± 10.34 0.214
Caregiver
Perceived
Stress
Perceived Stress 30.2 ± 8.2 33.2 ± 10.3 30.4 ± 7.7 33.4 ± 9.6 §0.100
§ in front of p value indicates variables chosen for multivariable analysis
ANTECEDENTS MARIJUANA INITIATION
43
Table 3.1: Multinomial Logistic Regression of Socio-demographic predictors (Early Adolescence vs Never Initiated)
Early Adolescence (14-15 Yrs)
Model 1 Model 2 Model 3
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
Wave 1 Age 1.69 (0.98, 2.92) 1.77 (1.00, 3.12) *2.21 (1.06, 4.63) 14%
Gender
(Ref=Male)
Female 0.77 (0.43, 1.37) 1.08 (0.52, 2.25) 1.35 (0.50, 3.64)
Race (ref=white)
Hispanic 1.59 (0.79, 3.21) 1.08 (0.45, 2.58) 0.33 (0.10, 1.08)
African
American
0.76 (0.27, 2.17) 0.56 (0.17, 1.87) 0.28 (0.05, 1.44)
Asian 0.21 (0.03, 1.72) 0.24 (0.03, 2.14) *0.06 (0.01, 0.79) 0%
Multiracial 0.48 (0.17, 1.32) 0.45 (0.14, 1.49) 0.25 (0.06, 1.17)
Zygosity
(ref=dizygotic)
Monozygotic *2.67 (1.37, 5.17) 57.9% *2.48 (1.15, 5.32) 46.8% *3.17 (1.16, 8.65) 55.4%
Neighborhood Characteristics
Score
1.58 (0.95, 2.63) 1.60 (0.93, 2.76) 1.35 (0.54, 3.37)
*p<0.05; Model 1: Unadjusted; Model 2: Adjusted for age, gender, race, Zygosity, and neighborhood characteristics score; Model 3: Adjusted for Model 2 +
Variables associated with attrition (Edinburg handedness, peer marijuana use, pregnancy planned?, how much did you want this pregnancy?, how happy was
your partner?, Hare Factor 1, Hare Factor 2, Junior Temperament & Character Inventory, CFQ: Overall Bullied scale, and Caregiver perceived stress).
ANTECEDENTS MARIJUANA INITIATION
44
Table 3.2: Multinomial Logistic Regression of Socio-demographic predictors (Mid-Adolescence vs Never Initiated)
Mid-Adolescence (16-18 Yrs)
Model 1 Model 2 Model 3
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
Wave 1 Age 1.53 (0.91, 2.56) 1.49 (0.86, 2.57) 1.67 (0.87, 3.19)
Gender
(Ref=Male)
Female *0.54 (0.31, 0.93) 10.7% 0.55 (0.29, 1.06) *0.38 (0.16, 0.89) 18.5%
Race (ref=white)
Hispanic 1.13 (0.58, 2.20) 0.87 (0.38, 1.98) 0.55 (0.19, 1.62)
African
American
0.66 (0.25, 1.73) 0.66 (0.22, 1.95) 0.55 (0.14, 2.18)
Asian
Multiracial 0.76 (0.35, 1.69) 0.63 (0.24, 1.67) 0.39 (0.12, 1.26)
Zygosity
(ref=dizygotic)
Monozygotic 1.14 (0.66, 1.96) 0.97 (0.50, 1.90) 1.43 (0.63, 3.25)
Neighborhood Characteristics
Score
1.09 (0.64, 1.87) 1.07 (0.60, 1.89) 1.39 (0.67, 2.88)
*p<0.05; Model 1: Unadjusted; Model 2: Adjusted for age, gender, race, Zygosity, and neighborhood characteristics score; Model 3: Adjusted for Model 2 +
Variables associated with attrition (Edinburg handedness, peer marijuana use, pregnancy planned?, how much did you want this pregnancy?, how happy was
your partner?, Hare Factor 1, Hare Factor 2, Junior Temperament & Character Inventory, CFQ: Overall Bullied scale, and Caregiver perceived stress).
ANTECEDENTS MARIJUANA INITIATION
45
Table 3.3: Multinomial Logistic Regression of Socio-demographic predictors (Late Adolescence vs Never Initiated)
Late Adolescence (19-20 Yrs)
Model 1 Model 2 Model 3
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
Wave 1 Age 1.33 (0.64, 2.77) 1.32 (0.62, 2.80) 1.68 (0.67, 4.21)
Gender
(Ref=Male)
Female 0.62 (0.28, 1.40) 0.63 (0.25, 1.58) 0.55 (0.16, 1.89)
Race (ref=white)
Hispanic 0.54 (0.20, 1.48) 0.49 (0.15, 1.53) 0.18 (0.03, 1.02)
African
American
0.72 (0.21, 2.46) 0.49 (0.11, 2.06) 1.20 (0.17, 8.65)
Asian
Multiracial 0.45 (0.14, 1.50) 0.52 (0.14, 1.91) 0.26 (0.04, 1.75)
Zygosity
(ref=dizygotic)
Monozygotic *2.84 (1.10, 7.35) 29.6% 2.34 (0.84, 6.52) 1.62 (0.45, 5.76)
Neighborhood Characteristics
Score
0.63 (0.25, 1.59) 0.72 (0.28, 1.89) 0.40 (0.10, 1.69)
*p<0.05; Model 1: Unadjusted; Model 2: Adjusted for age, gender, race, Zygosity, and neighborhood characteristics score; Model 3: Adjusted for Model 2 +
Variables associated with attrition (Edinburg handedness, peer marijuana use, pregnancy planned?, how much did you want this pregnancy?, how happy was
your partner?, Hare Factor 1, Hare Factor 2, Junior Temperament & Character Inventory, CFQ: Overall Bullied scale, and Caregiver perceived stress).
ANTECEDENTS MARIJUANA INITIATION
46
Table 4.1: Multinomial Logistic Regression of Predictors (Early Adolescence vs Never Initiated)
Early Adolescence (14-15 Yrs)
Model 1 Model 2 Model 3
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
Twin Sip of Alcohol *2.53 (1.17, 5.49) 58.4% *2.53 (1.12, 5.72) 24.4% *3.83 (1.37, 10.75) 17.4%
Peer Activities:
Have Drunk Alcohol
3.23 (0.44, 23.64) 2.77 (0.36, 21.42) 3.33 (0.35, 32.02)
Prenatal Support:
Pregnancy Planned?
(ref=No)
Some 0.93 (0.25, 3.50) 0.53 (0.12, 2.32) 0.54 (0.10, 2.97)
Yes *0.37 (0.17, 0.83) 29.8% *0.22 (0.09, 0.58) 74.3% *0.15 (0.04, 0.48) 51.0%
Prenatal Support:
How Happy was Your Partner?
0.85 (0.60, 1.19) 0.79 (0.55, 1.14) 0.73 (0.48, 1.12)
Prenatal Support:
How Supportive was Your
Family?
*0.60 (0.41, 0.86) 58.1% *0.62 (0.41, 0.92) 48.2% 0.65 (0.40, 1.07)
Martial Satisfaction:
Overall how happy are you in
your relationship?
1.14 (0.75, 1.74) 1.04 (0.66, 1.64) 1.19 (0.65, 2.16)
Parent to Child Affect:
Parent Report-Negative
1.46 (0.83, 2.57) 1.53 (0.84, 2.79) *2.27 (1.01, 5.13) 16.8%
CFQ: Overall Friendship Scale *2.15 (1.39, 3.33) 71.6% *2.53 (1.55, 4.14) 80.5% *3.12 (1.64, 5.95) 80.1%
CFQ: Overall Bullied Scale *1.87 (1.18, 2.98) 56.5% *1.77 (1.07, 2.95) 39.5% 1.92 (0.96, 3.83)
CFQ: Number of Close Friends 1.07 (0.94, 1.22) 1.07 (0.94, 1.23) 1.05 (0.89, 1.24)
Caregiver Perceived Stress *1.04 (1.00, 1.08) 16.6% 1.04 (0.99, 1.08) 1.01 (0.96, 1.06)
*p<0.05; Model 1: Unadjusted; Model 2: Adjusted for age, gender, race, Zygosity, and neighborhood characteristics score; Model 3: Adjusted for Model 2 +
Variables associated with attrition (Edinburg handedness, peer marijuana use, pregnancy planned?, how much did you want this pregnancy?, how happy was
your partner?, Hare Factor 1, Hare Factor 2, Junior Temperament & Character Inventory, CFQ: Overall Bullied scale, and Caregiver perceived stress).
ANTECEDENTS MARIJUANA INITIATION
47
Table 4.2: Multinomial Logistic Regression of Predictors (Mid-Adolescence vs Never Initiated)
Mid-Adolescence (16-18 Yrs)
Model 1 Model 2 Model 3
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
Twin Sip of Alcohol 1.75 (0.81, 3.75) 1.67 (0.76, 3.69) 1.78 (0.68, 4.63)
Peer Activities:
Have Drunk Alcohol
*9.21 (1.85, 45.89) 0.70% *7.93 (1.56, 40.30) 0.70% *7.31 (1.17, 45.70) 0.10%
Prenatal Support:
Pregnancy Planned?
(ref=No)
Some 0.88 (0.24, 3.23) 0.71 (0.18, 2.88) 0.60 (0.11, 3.36)
Yes 0.61 (0.31, 1.23) *0.43 (0.19, 0.97) 68.9% 0.46 (0.17, 1.25)
Prenatal Support:
How Happy was Your Partner?
0.84 (0.61, 1.15) 0.85 (0.61, 1.18) 0.88 (0.60, 1.29)
Prenatal Support:
How Supportive was Your
Family?
0.75 (0.52, 1.08) 0.75 (0.51, 1.09) 0.80 (0.51, 1.25)
Martial Satisfaction:
Overall how happy are you in
your relationship?
0.91 (0.61, 1.35) 0.94 (0.62, 1.42) 0.95 (0.58, 1.54)
Parent to Child Affect:
Parent Report-Negative
*1.91 (1.13, 3.25) 17.3% *2.13 (1.21, 3.74) 22.0% 2.05 (0.99, 4.24)
CFQ: Overall Friendship Scale *1.45 (1.00, 2.10) 60.6% *1.54 (1.03, 2.32) 62.4% *1.69 (1.03, 2.78) 61.1%
CFQ: Overall Bullied Scale 1.54 (0.99, 2.41) 1.59 (0.98, 2.56) 1.42 (0.77, 2.62)
CFQ: Number of Close Friends 1.06 (0.95, 1.20) 1.06 (0.94, 1.20) 1.11 (0.96, 1.27)
Caregiver Perceived Stress 1.00 (0.97, 1.04) 1.02 (0.98, 1.06) 0.99 (0.94, 1.04)
*p<0.05; Model 1: Unadjusted; Model 2: Adjusted for age, gender, race, Zygosity, and neighborhood characteristics score; Model 3: Adjusted for Model 2 +
Variables associated with attrition (Edinburg handedness, peer marijuana use, pregnancy planned?, how much did you want this pregnancy?, how happy was
your partner?, Hare Factor 1, Hare Factor 2, Junior Temperament & Character Inventory, CFQ: Overall Bullied scale, and Caregiver perceived stress).
ANTECEDENTS MARIJUANA INITIATION
48
Table 4.3: Multinomial Logistic Regression of Predictors (Late Adolescence vs Never Initiated)
Late Adolescence (19-20 Yrs)
Model 1 Model 2 Model 3
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
OR (95%CI)
% Effect
Replication
Twin Sip of Alcohol 1.85 (0.65, 5.24) 1.82 (0.62, 5.35) 1.58 (0.39, 6.33)
Peer Activities:
Have Drunk Alcohol
3.07 (0.27, 35.39) 2.86 (0.24, 34.36) 1.62 (0.10, 26.15)
Prenatal Support:
Pregnancy Planned?
(ref=No)
Some 0.56 (0.06, 5.04) 0.30 (0.03, 2.94) 0.20 (0.01, 2.63)
Yes 0.67 (0.25, 1.77) 0.35 (0.11, 1.10) 0.23 (0.05, 1.11)
Prenatal Support:
How Happy was Your Partner?
1.01 (0.63, 1.62) 0.96 (0.59, 1.56) 0.91 (0.52, 1.60)
Prenatal Support:
How Supportive was Your
Family?
0.78 (0.47, 1.29) 0.76 (0.45, 1.27) 0.86 (0.45, 1.63)
Martial Satisfaction:
Overall how happy are you in
your relationship?
*1.87 (1.19, 2.93) 10.6% *2.34 (1.37, 3.99) 11.8% *2.59 (1.21, 5.58) 25.0%
Parent to Child Affect:
Parent Report-Negative
1.83 (0.88, 3.82) 1.91 (0.90, 4.05) 2.11 (0.80, 5.55)
CFQ: Overall Friendship Scale 1.51 (0.89, 2.54) 1.46 (0.82, 2.57) 1.22 (0.58, 2.58)
CFQ: Overall Bullied Scale 1.69 (0.95, 3.00) *1.89 (1.02, 3.51) 0.10% 1.88 (0.77, 4.60)
CFQ: Number of Close Friends 1.14 (0.97, 1.34) 1.13 (0.95, 1.35) 1.08 (0.88, 1.34)
Caregiver Perceived Stress 1.04 (0.99, 1.10) *1.06 (1.01, 1.12) 1.10% *1.09 (1.01, 1.16) 1.56%
*p<0.05; Model 1: Unadjusted; Model 2: Adjusted for age, gender, race, Zygosity, and neighborhood characteristics score; Model 3: Adjusted for Model 2 +
Variables associated with attrition (Edinburg handedness, peer marijuana use, pregnancy planned?, how much did you want this pregnancy?, how happy was
your partner?, Hare Factor 1, Hare Factor 2, Junior Temperament & Character Inventory, CFQ: Overall Bullied scale, and Caregiver perceived stress).
ANTECEDENTS MARIJUANA INITIATION
49
Figures
Figure 1: Multivariable Multinomial Logistic Regression
Figure 2: Cross-Validation to Determine Interaction
0.01 0.10 1.00 10.00 100.00
Odds Ratio (Log Scale)
Alcohol Use-Twin
Alcohol Use-Peer
Prenatal Support-Parent
Relationship with Partner-Parent
Relationship with Twin-Parent
Friendships & Bullying-Twin
Percieved Stress-Parent
1.3
1.32 1.34 1.36 1.38
10-Fold Cross Validated Generalization Error
0 500 1000 1500 2000 2500 3000
Tree Iteration Number
Depth=1 Depth=2 Depth=3
Depth=4 Depth=5 Depth=6
Depth=7
Early Adolescence (14-15 Years)
Mid-Adolescence (16-18 Years)
Late Adolescence (19-20 Years)
ANTECEDENTS MARIJUANA INITIATION
50
Figure 3: Cross-Validation to Determine Tree Size
F i gu r e 4: R el a ti v e I nf l ue nc e of P r ed i c t ors of “ E v er Us e” f r o m G B M
1523
1.34 1.35 1.36 1.37 1.38
10-Fold Cross Validated Generalization Error
0 500 1000 1500 2000 2500 3000
Tree Iteration Number
0 1 2 3 4 5
Relative Influence of Variables
Demographic
Friendships & Bullying-Twin
Psychopathy-Twin
Temperament(JTCI)-Twin
Substance Use-Twin
Substance Use-Peer
Prenatal Substance Use-Parent
Relationship with Partner-Parent
Twin-Parent Relationship
Perceived Stress-Parent
Prenatal Support-Parent
ANTECEDENTS MARIJUANA INITIATION
51
Figure 5: ROC Curve of GBM Predictions
0.0 0.2 0.4 0.6 0.8 1.0
Sensitivity
0.0 0.2 0.4 0.6 0.8 1.0
1-Specificty (False Negative Rate)
Exploratory AUC=0.91 (0.86, 0.95)
Testing AUC=0.71 (0.64, 0.77)
ANTECEDENTS MARIJUANA INITIATION
52
Methods Supplement
Measures
Predictors of marijuana use. Sixty-one variables collected at Wave 1 from the twin and
parent, prior to any twin reporting marijuana use were used to predict future marijuana use.
These measures were: Socio-demographic variables such as age at Wave 1, gender, race,
Zygosity, socioeconomic status (based on Hollingshead ratings), handedness, and neighborhood
characteristics score; Substance-Use variables such as twin-reported cigarette or alcohol use,
peer cigarette, alcohol, marijuana, or drug use, and mother’s cigarette, alcohol, or drug use
during pregnancy; Family & Peer Relationship variables such as parent and twin-report on
positive/negative parent-to-child affect, parent’s relationship with their partner, and variables
pertaining to twin friendships and bully victimization; Parental Psychosocial variables such as
perceived stress and prenatal support; and Personality variables such as factors from the Junior
Temperament and Character Inventory (JTCI) (Luby et al., 1999) and the Childhood
Psychopathy Scale (CPS) (Lynam, 1997).
Socio-demographic. Socioeconomic status was calculated as a composite score based on
the Hollingshead 9-point rating scale containing information about the parent’s education,
occupation, and income. Handedness was evaluated through the Edinburgh Handedness
Inventory (Oldfield, 1971). The neighborhood characteristics score was calculated as the mean of
17 items that asked about how frequent problems such as gangs, prostitution, unemployment etc.
were in the respondent’s neighborhood. Item responses were on a 5-point Likert scale ranging
from Never to Always. A copy of the neighborhood scale is available in the Questionnaire
Appendix.
ANTECEDENTS MARIJUANA INITIATION
53
Substance use. Twins reported on their own use of cigarettes (puff of a cig) and alcohol
(sip, more than a sip, and whole drink) through the Substance Use Questionnaire. Peer substance
use was reported by the twin through the Peer Activities questionnaire (PA) and the Peer
Delinquency Interview (PDI). The PA asked “How many of your friends have….” smoked
cigarettes, drunk alcohol, used marijuana, used other drugs. Responses were on a 5-point scale of
None(1), A few (2), About Half(3), Most(4), and All(5). The PDI was the only questionnaire
used that was collected from Wave 2 of the study (ages 12-14). The PDI similarly asked “How
many of your friends have…” drank any alcohol, drank more than a sip, had a whole drink, tried
marijuana, tried other drugs. Responses were on a 4-point scale of None(1), 1-2 Friends(2), 3-4
Friends(3), and 5+ Friends(4). Mother’s substance use during pregnancy was assessed using
items from the Maternal Health Questionnaire (MHQ) which asked if the mother smoked
cigarettes, drank alcohol, or used drugs during their pregnancy.
Family & peer relationships. The Twin’s relationship with their primary caregiver was
assessed using the Parent-to-Child Affect questionnaire (PCA) which consisted of a 25 item
positive affect scale and 5 item negative affect scale (see appendix). Both parent and twin reports
were used as potential predictors with the Pearson correlation between parent and twin report at
0.23 for positive affect and 0.15 for negative (p’s < 0.001). The parent’s relationship with their
partner was assessed using items from the Marital Satisfaction survey (MS). The five items from
the MS asked the caregiver to rate their relationship happiness with respect to 1) time spent
together 2) effort partner puts into the relationship 3) amount of attention and affection 4) sexual
life/intimacy satisfaction and 5) overall happiness. Response options were on a 5-point Likert
scale ranging from Very Happy(1) to Very Unhappy(5). Twin friendships and bully
victimization were examined using the Childhood Friendship Questionnaire (CFQ). The CFQ is
ANTECEDENTS MARIJUANA INITIATION
54
non-standard questionnaire comprised of multiple items that asks about the twin’s level of
friendships (friendship initiator scale and friendship receiver scale), bully victimization at school
(physical scale, verbal scale, and indirect scale) and in the neighborhood, as well as a scale
pertaining to parental disapproval of friends. The items comprising each scale are recorded on a
5-point Likert response ranging from Never(1) to Almost Always(5). The number of causal and
close friends the twin had was also recorded. Details about the items comprising each CFQ scale
and the internal consistency of these scales can be found in the appendix.
Parental psychosocial. The caregiver’s perceived stress scale was measured as the mean
of 13 items with responses on a 5-point Likert scale ranging from Hardly Ever(1) to Always(5).
A copy of the perceived stress scale is available in the appendix. The primary caregiver was
typically the mother (~91%) or father (~6%). Prenatal Support was assessed by 6 separate items
1) was the pregnancy was panned; 2) how depressed were they when they found out they were
pregnant; 3) how much did they want the pregnancy; 4) how happy was their partner about the
pregnancy; 5) how supportive was their family about the pregnancy; 6) were they depressed after
the birth. Response options were on a 5-point Likert scale for items 2-6, with values ranging
from Not at All(1) to Very Much(5). Item 1 response options were Yes, Somewhat, and No.
Personality. The Junior Temperament and Character Inventory (JTCI) is an extension of
Cloninger’s TCI that has been adapted for children (Luby et al., 1999). In this sample, factors
pertaining to temperament (novelty seeking, harm avoidance, reward dependence, and
persistence) and character (self-directed and cooperative) were used as predictors. The
Childhood Psychopathy Scale (CPS) Revised Extended version (Lynam, 1997) was used to
assess the twin’s psychopathic personality traits. For this study, Hare Factor 1 (shallow affect,
superficial charm, manipulativeness, lack of empathy) and Factor 2 (criminal versatility,
ANTECEDENTS MARIJUANA INITIATION
55
impulsiveness, irresponsibility, poor behavior controls, juvenile delinquency) based on Hare’s
psychopathy checklist were utilized (Hare, 1991). Two additional factors pertaining to
callousness-disinhibited and manipulative-deceitful traits from this data were used and have been
reported on elsewhere (Bezdjian, Raine, Baker, & Lynam, 2011).
ANTECEDENTS MARIJUANA INITIATION
56
Supplementary Tables
Supplementary Table S1: Attrition Analyses
Variable Type Variable Name Values
Complete
Attrition
N=173
Partial
Attrition
N=254
Partial-
Attrition
with
Inference
N=184
Complete
Cases
N=167
P
Socio-
Demographics
Age 9.61 ± 0.59 9.54 ± 0.55 9.53 ± 0.59 9.74 ± 0.61 <.05
Gender
1:M 54.34% 52.36% 46.20% 44.31%
NS
2:F 45.66% 47.64% 53.80% 55.69%
Race
1: Cau 21.97% 33.47% 29.12% 31.74%
NS
2: His 37.57% 32.27% 36.81% 32.93%
3: AA 17.34% 12.35% 10.99% 13.17%
4: Asian 3.47% 3.98% 6.04% 2.99%
5: NatAmer 0.58% 0% 0% 0%
7: Multi 19.08% 17.93% 17.03% 19.16%
Zygosity
0: MZ 47.98% 47.04% 32.61% 45.51%
<.01
1: DZ 52.02% 52.96% 67.39% 54.49%
Socioeconomic Status 40.6 ± 11.5 42.8 ± 12.0 42.2 ± 12.1 43.8 ± 12.8 NS
Neighborhood Characteristics Score 1.73 ± 0.67 1.65 ± 0.60 1.64 ± 0.66 1.57 ± 0.54 NS
Edinburg Hand Score 56.1 ± 51.4 65.9 ± 43.7 63.7 ± 50.3 65.9 ± 46.5 NS
Handedness %Right 85.98% 94.00% 93.33% 92.22% <.05
Twin
Substance Use
Puff of Cig %Yes 1.80% 1.69% 2.34% 0% NS
Sip of Alcohol %Yes 25.75% 24.72% 24.22% 19.53% NS
More than Sip %Yes 3.59% 1.12% 3.13% 1.56% NS
Whole Drink %Yes 0.77% 0.74% 0% 0% NS
Peer
Substance Use
PA: Have Smoked
Cigs
%Yes 4.76% 2.29% 4.00% 1.56% NS
PA: Have Drunk Alc %Yes 10.78% 5.11% 5.65% 3.94% NS
PA: Have used Mar %Yes 5.42% 1.71% 0.81% 1.56% <.05
PA: Have used Drugs %Yes 3.59% 1.14% 0.80% NS
PDI: Smoked Cigs 1.16 ± 0.59 1.10 ± 0.43 1.11 ± 0.41 1.12 ± 0.35 NS
PDI: Sip of Alcohol 1.21 ± 0.54 1.30 ± 0.66 1.32 ± 0.64 1.23 ± 0.59 NS
PDI: More than Sip 1.46 ± 0.88 1.55 ± 0.96 1.52 ± 0.93 1.41 ± 0.62 NS
PDI: Whole Drink 2.00 ± 1.22 1.92 ± 1.16 1.88 ± 0.99 1.38 ± 0.52 NS
PDI: Marijuana Use 1.09 ± 0.38 1.12 ± 0.52 1.04 ± 0.21 1.07 ± 0.29 NS
PDI: Drugs 1.04 ± 0.36 1.05 ± 0.32 1.07 ± 0.37 1.02 ± 0.14 NS
Mom
Substance Use
in Pregnancy
MHQ: Smoke Cigs %Yes 11.18% 4.68% 5.69% 7.87% NS
MHQ: Consume
Alcohol
%Yes 8.75% 8.98% 9.68% 11.02% NS
MHQ: Drug Use %Yes 3.11% 3.55% 4.00% 2.38% NS
Prenatal
Support
Pregnancy Planned
1: No 59.86% 44.72% 47.15% 41.67%
<.05 2: Some 2.11% 8.07% 8.94% 5.00%
3: Yes 38.03% 47.20% 43.90% 53.33%
How depressed when found out
preg?
1.95 ± 1.31 1.66 ± 1.10 1.80 ± 1.23 1.64 ± 1.17 NS
How much did you want this preg? 3.60 ± 1.47 4.09 ± 1.19 3.90 ± 1.25 4.03 ± 1.32 <.01
How happy was your partner? 3.78 ± 1.35 3.98 ± 1.22 3.86 ± 1.20 4.18 ± 0.92 <.05
How supportive was your family? 3.93 ± 1.27 4.18 ± 1.03 4.07 ± 1.07 4.24 ± 0.90 NS
Depression after birth? 1.91 ± 1.24 1.78 ± 1.21 1.91 ± 1.16 1.84 ± 1.14 NS
Marital
Satisfaction
Amount of Time Spent Together 2.23 ± 1.05 2.29 ± 1.12 2.37 ± 1.13 2.31 ± 0.97 NS
Amount of effort partner puts into
relationship
2.18 ± 1.10 2.31 ± 1.18 2.27 ± 1.15 2.19 ± 1.02 NS
ANTECEDENTS MARIJUANA INITIATION
57
Amount of attention partner gives
you
2.10 ± 1.07 2.12 ± 1.05 2.16 ± 1.07 2.09 ± 1.01 NS
Satisfied sex life/intimacy 2.15 ± 1.17 2.01 ± 1.05 2.08 ± 1.15 2.11 ± 0.97 NS
Overall, how happy are you in your
relationship
1.95 ± 1.05 2.00 ± 1.02 1.97 ± 1.02 1.81 ± 0.90 NS
Childhood
Psychopathy
Scale
Hare Factor 1:
(shallow affect, superficial charm,
manipulativeness, lack of empathy)
0.253 ± 0.125 0.218 ± 0.107 0.221 ± 0.122 0.199 ± 0.104 <.01
Hare Factor 2:
(criminal versatility, impulsiveness,
irresponsibility, poor behavior
controls, juvenile delinquency)
0.268 ± 0.152 0.247 ± 0.153 0.232 ± 0.147 0.214 ± 0.138 <.05
Bezdjian Factor 1: (callousness-
disinhibited)
0.290 ± 0.138 0.275 ± 0.142 0.258 ± 0.135 0.249 ± 0.137 NS
Bezdjian Factor 2: (manipulative-
decietful)
0.224 ± 0.142 0.181 ± 0.120 0.189 ± 0.137 0.156 ± 0.110 <.01
Junior
Temperament
(T) and
Character (C)
Inventory
Novelty Seeking (T) 0.268 ± 0.162 0.255 ± 0.160 0.244 ± 0.153 0.217 ± 0.125 <.05
Harm Avoidance (T) 0.533 ± 0.186 0.495 ± 0.197 0.494 ± 0.197 0.470 ± 0.194 <.05
Reward Depend (T) 0.483 ± 0.185 0.501 ± 0.198 0.512 ± 0.205 0.543 ± 0.174 NS
Persistence (T) 0.644 ± 0.202 0.669 ± 0.213 0.702 ± 0.222 0.692 ± 0.213 NS
Self-Directed(C) 0.640 ± 0.168 0.674 ± 0.169 0.688 ± 0.163 0.715 ± 0.159 <.01
Cooperative (C) 0.746 ± 0.141 0.785 ± 0.142 0.805 ± 0.130 0.814 ± 0.137 <.01
Parent to Child
Affect
Parent Report-Positive 4.09 ± 0.46 4.13 ± 0.40 4.15 ± 0.42 4.13 ± 0.41 NS
Parent Report-Negative 2.34 ± 0.64 2.35 ± 0.59 2.28 ± 0.63 2.33 ± 0.59 NS
Twin Report-Positive 3.80 ± 0.63 3.85 ± 0.59 3.88 ± 0.65 3.99 ± 0.52 NS
Twin Report-Negative 2.29 ± 0.75 2.22 ± 0.66 2.15 ± 0.84 2.13 ± 0.70 NS
Child
Friendship
Questionnaire
Friendship Initiator Scale 3.76 ± 1.02 3.82 ± 1.01 3.63 ± 1.05 3.94 ± 1.05 NS
Friendship Receiver Scale 3.59 ± 0.96 3.47 ± 1.01 3.44 ± 0.97 3.71 ± 1.06 NS
Overall Friendship Scale 3.63 ± 0.81 3.58 ± 0.88 3.51 ± 0.88 3.78 ± 0.94 NS
Bullied @ School: Indirect (exile)
Scale
2.03 ± 1.11 2.16 ± 1.09 2.04 ± 1.12 1.96 ± 1.12 NS
Bullied @ School: Verbal Scale 1.96 ± 0.91 1.89 ± 0.97 1.91 ± 1.02 1.76 ± 0.93 NS
Bullied @ School: Physical Scale 1.68 ± 0.93 1.69 ± 1.04 1.50 ± 0.79 1.44 ± 0.84 NS
Bullied @ School: Overall Scale 1.87 ± 0.78 1.85 ± 0.88 1.79 ± 0.87 1.68 ± 0.72 NS
Bullied in Neighborhood Scale 1.83 ± 1.01 1.84 ± 0.96 1.69 ± 0.96 1.46 ± 0.72 <.01
Overall Bullied Scale 1.84 ± 0.76 1.83 ± 0.83 1.75 ± 0.85 1.62 ± 0.67 NS
Parental Disapproval of Friends
Scale
1.48 ± 0.58 1.50 ± 0.62 1.45 ± 0.57 1.37 ± 0.52 NS
# Of Close Friends 5.57 ± 2.93 6.12 ± 2.95 5.59 ± 2.72 5.73 ± 2.93 NS
# Of Casual Friends 9.66 ± 17.73 8.82 ± 15.33 8.66 ± 13.79 7.86 ± 13.27 NS
Caregiver
Perceived
Stress
Perceived Stress 33.1 ± 9.0 32.7 ± 8.9 31.9 ± 8.7 30.2 ± 8.6 <.05
ANTECEDENTS MARIJUANA INITIATION
58
Supplementary Table S2: Relative Influence of Predictors in GBM
Variable
Rel.
Inf.
Variable
Rel.
Inf.
Child Friendship: Overall Friendship Scale 5.016 Peer Delinquency Interview: More than Sip of Alcohol? 1.394
Child Friendship: Overall Bullied Scale 4.967 Child Friendship: Bullied @ School: Verbal Scale 1.351
Socioeconomic Status 3.909 JTCI: Persistence 1.351
Caregiver Perceived Stress 3.594 Marital Satisfaction: Satisfied sex life/intimacy 1.341
Parent to Child Affect: Twin Report-Positive 3.574 Prenatal Support: How much did you want this preg? 1.340
Childhood Psychopathy Scale: Bezdjian Factor 1 3.503 JTCI: Cooperative 1.315
Child Friendship: Parental Disapproval of Friends Scale 3.365 Parent to Child Affect: Twin Report-Negative 1.274
Child Friendship: Bullied @ School: Overall Scale 3.141 JTCI: Reward Depend 1.090
Male Gender 3.136 Childhood Psychopathy Scale: Bezdjian Factor 2 1.089
Peer Delinquency Interview: Smoked Cigs? 2.908 Peer Delinquency Interview: Sip of Alcohol? 0.999
Prenatal Support: How happy was your partner? 2.906 Child Friendship: Bullied @ School: Physical Scale 0.981
Edinburg Handedness Inventory 2.777 Twin Substance Use: Sip of Alcohol 0.940
Child Friendship: Friendship Receiver Scale 2.602 Marital Satisfaction: Amount of Time Spent Together 0.905
Neighborhood Characteristics Score 2.564
Marital Satisfaction: Overall, how happy are you in your
relationship
0.810
JTCI: Harm Avoidance 2.557
Marital Satisfaction: Amount of effort partner puts into
relationship
0.684
Age at Wave 1 2.489 Prenatal Support: Depression after birth? 0.552
Prenatal Support: Pregnancy Planned 2.463
Prenatal Support: How depressed when found out
pregnant?
0.359
Parent to Child Affect: Parent Report-Positive 2.297 Mom Substance Use in Pregnancy: Consume Alcohol 0.065
Child Friendship: Bullied in Neighborhood Scale 2.255 Peer Delinquency Interview: Marijuana Use? 0.021
Child Friendship: Friendship Initiator Scale 2.220 Mom Substance Use in Pregnancy: Drug Use 0
Childhood Psychopathy Scale: Hare Factor 1 2.173 Mom Substance Use in Pregnancy: Smoke Cigs 0
JTCI: Novelty Seeking 2.081 Peer Delinquency Interview: Whole Drink of Alcohol? 0
Child Friendship: # Of Casual Friends 2.058 Peer Delinquency Interview: Drug Use 0
Prenatal Support: How supportive was your family? 2.012 Peer Activities: Have Drunk Alcohol 0
JTCI: Self-Directed 1.873 Peer Activities: Have Smoked Cigarettes 0
Parent to Child Affect: Parent Report-Negative 1.809 Peer Activities: Have Used Drugs 0
Childhood Psychopathy Scale: Hare Factor 2 1.766 Peer Activities: Have Used Marijuana 0
Zygosity 1.655 Twin Substance Use: More than Sip of Alcohol 0
Child Friendship: # Of Close Friends 1.578 Twin Substance Use: Whole Drink of Alcohol 0
Marital Satisfaction: Amount of attention partner gives you 1.489 Twin Substance Use: Puff of Cigarette 0
Child Friendship: Bullied @ School: Indirect (exile) Scale 1.403
ANTECEDENTS MARIJUANA INITIATION
59
Supplementary Figures
Supplementary Figure S1: Patterns of Missing Data & Using Reasonable Inference
Have you ever tried marijuana?
Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Initiation? Ever Use?
Subject1 No No No Never No
Subject2 No No No Missing D/K D/K
Subject3 No No Yes Wave 3 Yes
Subject4 No No Missing Yes D/K Yes
Note: Arrows represent how missing values were inferred from observed responses. D/K means Don’t Know.
Supplementary Figure S2: GBM Predictors Association with Ever Use
ANTECEDENTS MARIJUANA INITIATION
60
Appendix
Neighborhood Characteristics Scale:
I am going to read you a list of problems. For each one, please tell me how often it is a problem in your neighborhood.
Never Rarely Some Often Always DK
1. How often is high unemployment (or a lot of people out of work) a problem in
your
2. neighborhood?
3. Different racial or cultural groups who do not get along with each other?
4. Vandalism and property damage?
5. How often is little respect for rules, laws and authority a problem in your
neighborhood?
6. Winos and junkies?
7. Prostitution?
8. How often are abandoned houses a problem in your neighborhood?
9. Sexual assaults or rapes?
10. Burglaries and thefts?
11. How often is gambling a problem in your neighborhood?
12. Run down and poorly kept buildings and yards?
13. Mafia or organized crime?
14. How often are assaults and muggings a problem in your neighborhood?
15. Gangs?
16. Homeless people or street people?
17. How often are drug use and drug dealing in the open a problem in your
neighborhood?
18. Selling of stolen goods?
ANTECEDENTS MARIJUANA INITIATION
61
Perceived Stress Scale:
Now I'm going to ask you about your feelings and thoughts DURING THE LAST MONTH.
Hardly Occasion- Some- Pretty Always DK
Ever ally times often
1. In the last month, have you been upset because of things that happened suddenly or
unexpectedly?
2. How often have you felt like you were unable to control the important things in your life?
3. How often have you felt nervous and stressed?
4. How often have you felt things were going your way?
5. In the last month, how often have you found that you could not deal with all of the things
that you had to do?
6. How often have you been able to keep the hassles and irritations in your life under
control?
7. How often have you felt that you were on top of things?
8. How often have you been angry about things that happened to you?
9. In the last month, how often were you behind with things you needed to do?
10. How often have you been able to control the way you spend your time?
11. How often do you worry about having enough money for housing, food, clothing, or other
essentials?
12. How often do you worry about having enough money to do fun things (with your family)?
13. How often have you had trouble paying your bills this past year?
ANTECEDENTS MARIJUANA INITIATION
62
Childhood Friendship Questionnaire Scales:
Scales Items Comprising Scale
Cronbach’s
Alpha
Friendship Initiator
Scale
1. How often do you play with someone at recess?
2. How often do you invite someone over to your house to play?
3. How often do you invite someone over to spend the night?
4. How often do you invite someone over for your birthday?
0.68
Friendship
Receiver Scale
1. How often does someone save you a seat at lunch?
2. How often does someone ask you to play at recess?
3. How often does someone invite you to their birthday party?
4. How often does someone else invite you over to their house to play?
5. How often does someone invite you over to spend the night?
0.72
Overall Friendship
Scale
Friendship Initiator Scale + Friendship Receiver Scale 0.80
Bullied @ School:
Indirect (exile)
Scale
1. How often does someone at school ignore you or refuse to talk to you?
2. How often does someone at school tell others not to talk to you?
3. How often does someone at school try to get you in trouble with your
friends or make your friends turn against you?
4. How often does someone at school tell you that you can't play with them
or join their group?
0.77
Bullied @ School:
Verbal Scale
1. How often does someone at school tease or make fun of you?
2. How often does someone at school gossip or say mean things about you behind your back?
3. How often does someone at school swear at you or say bad words to you?
4. How often does someone at school say they are going to hit or hurt you?
5. How often does someone at school threaten you with a weapon?
6. How often does someone at school call you names?
7. Ho w of ten do es s om eo ne a t s c ho ol tel l y ou y o u’ r e dumb or stupid?
8. How often does someone at school make fun of you because of the way you look?
9. How often does someone at school make mean remarks about the color of your skin?
10. How often does someone at school make fun of your religion or beliefs?
0.88
Bullied @ School:
Physical Scale
1. How often does someone at school hit, push, or start fights with you?
2. How often does someone at school try to break or damage something of yours?
3. How often does someone at school hurt you physically somehow?
4. How often does someone at school mess up your clothing in some way?
5. How often does someone at school take something of yours without your permission?
6. How often does someone at school steal money from you?
7. How often does someone at school pinch you?
8. How often does someone at school punch you?
9. How often does someone at school kick you?
0.85
Bullied @ School:
Overall Scale
Bullied @ School: Physical Scale + Bullied @ School: Verbal Scale +
Bullied @ School: Indirect (exile) Scale
0.93
ANTECEDENTS MARIJUANA INITIATION
63
Bullied in
Neighborhood
Scale
1. How often does someone play mean tricks on you when you are NOT AT SCHOOL?
2. How often does someone tell you that you can't play with them or join their group when you are
NOT AT SCHOOL?
3. How often does someone tease you or make fun of you when you are NOT AT SCHOOL?
0.76
Overall Bullied
Scale
Bullied in Neighborhood Scale + Bullied @ School: Overall Scale 0.94
Parental
Disapproval of
Friends Scale
1. How many of your friends does your mother dislike or disapprove of?
2. How many of your friends does your father dislike or disapprove of?
3. How many of your friends does your mother think are a bad influence on you?
4. How many of your friends does your father think are a bad influence on you?
5. How many of your friends do things that your mother would NOT approve of?
6. How many of your friends do things that your father would NOT approve of?
0.81
ANTECEDENTS MARIJUANA INITIATION
64
Parent-to-Child Affect Scale
(P=Item from Positive Affect scale; N=Item from Negative Affect Scale) Never Rarely Some Often Always DK
1. (P) How often does your caregiver hug or kiss you?
2. (P) How often do you spend time alone with your caregiver?
3. (N) How often do you think your caregiver wants you to stop bothering him/her, and leave
him/her alone for a few minutes?
4. (P) How often does your caregiver say, " I love you" to you?
5. (P) How often does your caregiver help you with schoolwork?
6. (N) How often does your caregiver nag you about what you are doing wrong?
7. (P) How often does your caregiver tell you that you're special?
8. (P) How often does your caregiver play games or sports with you?
9. (N) How often does your caregiver criticize you?
10. (P) How often does your caregiver praise you for something you've done?
11. (P) How often do you go for walks with your caregiver?
12. (P) How often does your caregiver accept you as you are?
13. (P) How often does your caregiver tell you you're a "good boy" or "good girl?"
14. (P) How often does your caregiver get involved in your activities (including sports, music,
scouts, etc.)?
15. (N) How often do you think your caregiver wishes you were different in some ways?
16. (P) Is your caregiver a good listener?
17. (P) Can your caregiver tell how you are feeling without asking?
18. (P) If you were in trouble, could you tell your caregiver?
19. (P) Does your caregiver try to understand your point of view?
20. (N) Are there things you avoid talking about with your caregiver?
21. (P) Do you find it easy to discuss problems with your caregiver?
22. (P) Is it very easy for you to express your feelings to your caregiver?
23. (P) Can you count on your caregiver for help with things like homework or rides when you need
it?
24. (P) How well do you and your caregiver communicate with each other?
25. (P) How well does your caregiver understand you?
26. (P) How close are you to your caregiver?
27. (P) How well can you trust your caregiver with your secrets?
28. (P) How well can you count on your caregiver for comfort when you're hurt or sad?
29. (P) How much do you think you are like your caregiver?
30. (P) How happy are you about the amount of time you spend with your caregiver?
Abstract (if available)
Abstract
Marijuana use among adolescents has been increasing and is associated with loss in cognitive abilities, risky behavior, and later substance abuse and dependence. Specifically, early initiation is associated with long-term negative outcomes. Data from The University of Southern California (USC) Risk Factors for Antisocial Behavior (RFAB) twin study were used to examine the psychosocial antecedents of marijuana initiation in adolescence. Participants were assessed in Wave 1 (ages 9-10) for risk and resiliency factors prior to ever having used marijuana. Marijuana use was assessed every 2-3 years, with the participants currently 19-20 years old (Wave 5-ongoing). The sample was randomly split such that one twin was assigned to an exploratory dataset and the co-twin to a testing (or confirmation) dataset. The exploratory dataset was used for model building and the testing dataset was used to determine the robustness or accuracy of the model. Wave 1 predictors were used in multinomial logistic regression models to predict initiation age groups. Gradient Boosted Modeling was used to identify influential predictors of ever having used marijuana. Predictors of initiation were related to friendships and bully victimization, sociodemographics, parental stress, and the parent-child relationship.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Multivariate biometric modeling among multiple traits across different raters in child externalizing behavior studies
PDF
Different genetic and environmental structures for the overlap of three antisocial behavior factors with alcohol initiation
PDF
The design, implementation, and evaluation of accelerated longitudinal designs
PDF
A comparison of standard and alternative measurement models for dealing with skewed data with applications to longitudinal data on the child psychopathy scale
PDF
Attrition in longitudinal twin studies: a comparative study of SEM estimation methods
PDF
Sources of stability and change in the trajectory of openness to experience across the lifespan
PDF
Patterns of EEG spectral power in 9-10 year old twins and their relationships with aggressive and nonaggressive antisocial behavior in childhood and adolescence
PDF
Family conflict, negative mood, and adolescents' daily problems in school
PDF
Adolescent conduct problems and substance use: an examination of the risk pathway across the transition to high school
PDF
A functional use of response time data in cognitive assessment
PDF
Evaluating social-cognitive measures of motivation in a longitudinal study of people completing New Year's resolutions to exercise
PDF
An examination of in-school and online protective factors for adolescent trajectories of online victimization and harassment over time
PDF
Biometric models of psychopathic traits in adolescence: a comparison of item-level and sum-score approaches
PDF
Factors and correlates of sexual behaviors among young adults from continuation high schools
PDF
A penta-dimensional longitudinal analysis of the predictors of compulsive internet use among adolescents using linear mixed model (LMM)
PDF
Psychophysiological reactivity to stress, psychopathic traits, and developmental trajectories of antisocial behavior: a prospective, longitudinal study
PDF
Developmental etiology of reactive and proactive aggression, electrodermal reactivity and their relationships: a twin study
PDF
A longitudinal study of sexual minority stress and behavioral health in sexual minority adolescents
PDF
Later life success of former college student-athletes as a function of retirement from sport and participant characteristics
PDF
Tobacco and marijuana surveillance using Twitter data
Asset Metadata
Creator
Jackson, Nicholas J.
(author)
Core Title
Antecedents of marijuana initiation
School
College of Letters, Arts and Sciences
Degree
Master of Arts
Degree Program
Psychology
Publication Date
10/07/2014
Defense Date
10/07/2014
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
adolescence,gradient boosting,machine learning,marijuana initiation,OAI-PMH Harvest,statistical learning,substance use,twin study
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Baker, Laura A. (
committee chair
), McArdle, John J. (
committee member
), Monterosso, John R. (
committee member
)
Creator Email
njacks@gmail.com,njjackso@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-488477
Unique identifier
UC11287033
Identifier
etd-JacksonNic-3004.pdf (filename),usctheses-c3-488477 (legacy record id)
Legacy Identifier
etd-JacksonNic-3004.pdf
Dmrecord
488477
Document Type
Thesis
Format
application/pdf (imt)
Rights
Jackson, Nicholas J.
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
gradient boosting
machine learning
marijuana initiation
statistical learning
substance use
twin study