Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Research Article
Ved
ABSTRACT
Objective:
Due to little knowledge regarding the contextual factors related to cannabis use, we aimed to
provide descriptive statistics regarding contextual factors related to use and examine the predictive ability
of contextual factors.
Method:
We included college student participants (
n
= 5700; male = 2893, female =
3702, other gender identity = 48, missing = 57) from three multi-site studies in our analyses. We examined
the means and standard deviations of contextual factors related to cannabis use (social context/setting,
form of cannabis, route of administration, source of purchase, and proxies of use). Additionally, we tested
the predictive ability of the contextual factors on cannabis use consequences, protective behavioral
strategies, and severity of cannabis use disorder, via an exploratory machine learning model (random
forest).
Results:
Descriptive statistics and the correlations between the contextual factors and the three
outcomes are provided. Exploratory random forests indicated that contextual factors may be helpful in
predicting consequences and protective behavioral strategies and especially useful in predicting the
severity of cannabis use disorder.
Conclusions:
Contextual factors of cannabis use warrants further
exploration, especially considering the difficulty in assessing dosage when individuals are likely to consume
in a group context. We propose considering measuring contextual factors along with use in the past 30 days
and consequences of use.
Key words
: = cannabis use; cannabis-related consequences; social context; route of administration; college
students; cannabis protective behavioral strategies
In the context of a massively growing legal
cannabis market throughout the United States, a
harm reduction approach to understanding
cannabis necessitates the consideration of any
relevant characteristic of one’s cannabis use that
may contribute to cannabis-related harms. In a
meta-analysis, Pearson (2019) found a medium-
sized association between cannabis use indicators
and consequences (rw=.367), demonstrating that
most of the variance in cannabis-related negative
consequences are not explained by any single
indicator of cannabis use. This finding suggests
that additional characteristics of cannabis use are
needed to account for the likelihood of
experiencing cannabis-related harms beyond
frequency and quantity of use. Social contexts of
use, or the temporal, motivational, and situational
factors surrounding use, are additional
characteristics of cannabis use that predict
cannabis use outcomes (Beck et al., 2009).
Matison W. McCool1, Matthew R. Pearson1, the Marijuana
Outcomes Study Team*, the Protective Strategies Study Team**, &
the Addictions Research Team***
1Center on Alcohol, Substance Use, and Addictions, University of New Mexico
Cannabis
2024
© Author(s) 2024
researchmj.org
10.26828/cannabis/2024/000225
Beyond Frequency and
Quantity of Cannabis
Consumption among College
Students: Context of Using
Cannabis Relates to Cannabis-
Related Outcomes
Corresponding Author: Matison McCool, Ph.D., University of New Mexico, 2650 Yale Blvd SE MSC 11-6280.
Albuquerque, NM 87106. Phone: (505) 925-2322. E-mail: mwmccool@unm.edu.
Cannabis, A Publication of the Research Society on Marijuana
Contextual Factors of Cannabis Use
The operationalization of contextual factors of
cannabis use in prior work has included categories
related to social facilitation, peer acceptance, sex-
seeking, emotional pain (Beck et al., 2009),
location of use, using companions (Spinella et al.,
2019), place of purchase, route of administration
(Parnes et al., 2018), environmental, emotional,
and interpersonal contexts related to use (Gray et
al., 2024). Recalling that cannabis use indicators
such as quantity and frequency are moderately
associated with consequences (Pearson, 2019),
contextual factors are associated with cannabis
use disorder (Beck et al., 2009), consequences, and
protective behavioral strategy use (Gray et al.,
2024) even when controlling for direct cannabis
use indicators. However, much of the previous
literature has limited the examination of
contextual factors to a few behaviors (e.g., solitary
use; Spinella et al., 2019), creating constructs out
of specific behaviors (Beck et al., 2009), or using
latent profile analyses to find patterns of
contextual factors (Gray et al., 2024). Contextual
factors of cannabis use are significantly correlated
(Beck et al., 2009) which may lead to issues of
multicollinearity when attempting to examine
many individual contexts in a predictive model.
Given the broad operational definition of cannabis
use contextual factors and many distinct
situations in which cannabis use can occur, we
focused our study on the following contexts of
cannabis use: social situations and settings of use,
form of cannabis and route of administration, and
the source of purchase.
Situational and Setting Contexts
Most of the research on social context of
cannabis use has focused on solitary use vs. social
use. Compared to social users, solitary cannabis
users have reported higher levels of drinking to
cope, higher levels of cannabis use, and greater
endorsement of cannabis abuse/dependence
(Spinella et al., 2019). Solitary cannabis use by
adolescents has been shown to relate to cannabis
use disorder symptoms during adolescence, but
also prospectively predicts cannabis use disorder
symptoms in young adulthood (Creswell et al.,
2015). Solitary cannabis use frequency has been
shown to mediate (i.e., account for) the effects of
social anxiety on cannabis use and negative
cannabis-related consequences (Buckner et al.,
2016). Thus, solitary use of cannabis has been
identified as a risk factor for negative cannabis-
related consequences. Beck et al., (2019) included
settings of cannabis use (i.e., in a car, in a dorm
room) as part of a social facilitation construct.
Results assessing the relationship between social
facilitation and DSM-IV cannabis use disorder
criteria found that increased social facilitation
was significantly associated with cannabis use
disorder symptom severity. Overall, where and
with whom individuals use cannabis are
associated with cannabis use outcomes above and
beyond direct use indicators.
Context of Cannabis Form and Route of
Administration
With the rapid proliferation of legal cannabis
markets, cannabis preparations have diversified
to include a wide range of edible products and
high-concentration products, which have unique
routes of administration that are relevant to
cannabis-related harms (Parnes et al., 2018). For
example, oral ingestion of cannabis is associated
with higher concentrations of 11-hydroxy-∆9-
tetrahydrocannabinol (11-hydroxy-THC), which
may be more potent than ∆9-tetrahydrocannabinol
(THC) (Lemberger et al., 1973; Schwilke et al.,
2009), and may lead to delayed onset of
psychoactive effects, which leads to unintentional
overintoxication (we avoid using the term
overdose given that the primary intoxicating
chemical in cannabis is non-toxic and non-lethal).
High concentration products can be smoked with
an assortment of essential equipment but can also
be vaped in a concealable vape pen. Use of
concentrates is associated with rapid and higher
levels of intoxication compared to flower products
(Bidwell et al., 2020).
Source of Purchase Context
An outer situational context of one’s cannabis
use includes how one obtains cannabis products.
In the early days of recreational cannabis
legalization in Los Angeles (i.e., 2016-2017),
young adults who purchased products from
cannabis dispensaries (compared to obtaining
from family or friends) reported spending more
money on cannabis, using more distinct cannabis
products, using more frequently, using higher
Cannabis Use Context
quantities, using alone more often, and
experienced higher negative cannabis-related
consequences and cannabis use disorder
symptoms (D’Amico et al., 2020).
Brief Machine Learning Overview
Machine learning approaches differ from
traditional statistical approaches in several ways.
First, traditional statistical approaches have
focused on questions of inference, or using
probabilities to test hypotheses describing how
and why variables are related. Machine learning
algorithms largely focus on answering questions
related to prediction, or using existing data to find
patterns that predict a precise outcome (Bzdok et
al., 2018). While statistical models rely on
parametric assumptions about the relationship
between predictors and an outcome, machine
learning algorithms do not and look for complex
interactions to make the best prediction (Lantz,
2019; Witten & Frank, 2002). For example,
multiple regressions use independent variables,
as the predictors need to be independent from
each other to not affect the standard errors of
other predictors. As such, multicollinearity occurs
when an independent variable is highly correlated
with other independent variables resulting in
unstable coefficients and problems with model
convergence (Allen, 1997). Machine learning
models such as random forests are less affected by
correlated variables, as they do not attempt to
isolate the effects of a single variable on an
outcome when looking for complex interactions to
make predictions. Though, multi-collinearity can
slightly affect the selection of important variables
(Strobl et al., 2008).
Machine learning models offer unique
advantages in examining outcomes, specifically
regarding their ability to make precise
predictions. However, a trade-off exists such that
improved prediction is balanced by a loss in
explaining outcomes (inference) as no coefficients
are provided examining direct relationships
between predictors and outcomes. Machine
learning algorithms have been used to examine
cannabis-related outcomes such as consequences
from use (Schwebel et al., 2022), cannabis use in
daily life (Yu et al., 2023), and to examine the risk
and protective factors of cannabis use (Henry et
al., 2024).
The Present Study
Prior research has established relationships
between constructs or latent profiles of cannabis
contextual factors and cannabis protective
behavioral strategy use, cannabis use
consequences, and cannabis use disorder severity
(Beck et al., 2009; Dyar et al., 2021; Gray et al.,
2024; Parnes et al., 2018). However, grouping
contextual factors together through variable or
person-centered approaches limits the ability to
identify specific contexts that may be of
importance to predicting cannabis use outcomes.
We aimed to extend prior research by using
specific contextual indicators as predictors of
cannabis use outcomes within three large samples
of college student cannabis users. We sought to
broadly characterize the social context of cannabis
use among college students. We report descriptive
statistics across each sample, and then used an
exploratory modeling technique (random forest) to
identify salient contextual predictors related to
cannabis outcomes. Therefore, we examined
contextual factors as separate indicators of
cannabis protective behavioral strategies
(Pedersen et al., 2017), negative cannabis-related
consequences, and cannabis use disorder
symptoms.
METHODS
Participants and Procedure
The Marijuana Outcomes Study Team
(MOST) participants included college students
recruited from the psychology department
participant pools at 9 universities in 9 states
throughout the United States who participated for
research participation credit according to
procedures approved by the institutional review
boards at each participating university (for
methodological details regarding MOST please
see: Richards et al., 2021). Of 7,000 total
participants, our analyses are focused on 2,077
who reported past month cannabis use. Data were
collected between Fall 2016 and Spring 2017 such
that at the time of data collection two states
permitted recreational cannabis use (CO and
WA), 3 states permitted medical cannabis use
(NM, NY, and CA), and 4 states did not permit
cannabis use (VA, TX, TN, and FL).
Cannabis, A Publication of the Research Society on Marijuana
The Protective Strategies Study Team (PSST)
participants included college students recruited
using similar procedures from 10 universities in 10
states throughout the United States (for details
regarding PSST please see: Pearson et al., 2019). Of
7,303 total participants, our analyses are focused on
2,222 who reported past month cannabis use. Data
were collected between Spring 2017 and Fall 2017
such that at the time of data collection 3 states
permitted recreational cannabis use (AK, CO, and
WA), 1 state permitted medical cannabis use (NM),
and 6 states did not permit cannabis use (ID, MO,
MS, NE, VA, and WY).
The Addiction Research Team study (ART)
participants included college students recruited
using similar procedures from 10 universities in 8
states throughout the United States (for details
regarding the method including participants and
recruitment please see: Richards et al., 2022, 2023).
Of 5,594 total participants, our analyses are focused
on 1,397 who reported past month cannabis use.
Data were collected between Spring 2020 and Fall
2020 such that at the time of data collection 4 states
permitted recreational cannabis use (AK, CA, CO,
and WA), 1 state permitted medical cannabis use
(NM), and 3 states did not permit cannabis use (ID,
VA, TX). Participants in all studies provided
informed consent to participate.
In total, our analyses focused on 5700
participants (male = 2893, female = 3702, other
gender identity = 48, missing = 57). The average age
of the sample was 20.17 years (
SD
= 3.36). Most of
the participants identified as White (White = 4110,
American Indian/Alaska Native= 161, Asian = 568,
Black/African American= 861, Native
Hawaiian/Pacific Islander = 88, and Other = 432)
non-Hispanic (
n
= 4487).
Measures
Context of Cannabis Use.
MOST investigators
developed a broad assessment of contextual
variables related to one’s cannabis use to serve
various purposes. This assessment characterizes
the amount of money spent on cannabis; frequency,
level, and length of intoxication; social and physical
contexts of use; form of cannabis and route of
administration; level of unplanned use; and source
of cannabis (see Table 1 for the items, scales of
measurement, and descriptive statistics for these
items). Items in the context measure focused on
proxies for direct use (e.g., money spent, subjective
intoxication questions), social and setting places of
use (e.g., with friends, at home), form of cannabis
and route of administration (e.g., flower,
concentrate, using a bong, vaporizer), and source of
purchase (e.g., dispensary, black market). To focus
our analyses on the predictive ability of contextual
factors only, we excluded proxies of direct use. Most
scale items asked participants to rate the percent of
time they engaged in each contextual factor (0% -
100%). For example, participants were asked to
report the percentage of time they used each form of
cannabis, and totals had to equal 100%. Again,
please see Table 1 for the specific items and scales
of measurement regarding the context factors.
Table 1. Cannabis Use Contexts Across Datasets
MOST
PSST
ART
Total
[Variable labels are underlined]
M
SD
M
SD
M
SD
M
SD
Money Spent
(Please estimate how much money
you have spent on marijuana in the past month ($).)
42.44
69.57
45.20
70.95
53.38
76.02
46.19
71.84
Typical Intoxication
(On a typical marijuana use
day in the past 30 days, please indicate how high
you get from using marijuana (0 – 100%).)
60.77
24.41
61.64
25.14
64.09
23.13
61.84
24.41
Peak Intoxication
(Please indicate the highest you
have been from marijuana in the past month (0 –
100%).)
71.81
26.78
73.44
26.96
75.07
23.94
73.15
26.21
Peak Frequency
(What percentage of the time do
you get this high from using marijuana (0 – 100%)?)
57.75
32.09
59.27
32.10
62.48
30.59
59.49
31.77
Length of Intoxication
(On a typical marijuana use
day in the past 30 days, how long do you stay high
from using marijuana (hours)?)
3.83
13.86
3.27
5.09
2.87
2.02
3.39
8.99
Cannabis Use Context
MOST
PSST
ART
Total
[Variable labels are underlined]
M
SD
M
SD
M
SD
M
SD
Form of Cannabis
(In the past month, please report
the percentage of marijuana you consumed in each
of the following ways (must total to 100%))
Plant (i.e., bud, flower)
78.65
32.52
73.94
34.46
53.27
40.48
70.38
36.93
Edibles (i.e., brownie, chocolate)
10.79
23.97
12.97
26.54
17.21
30.72
13.26
26.96
Concentrates (i.e., hash, dabs)
8.62
21.46
11.38
23.84
26.07
35.90
14.07
27.60
Other [other form]
2.15
13.35
1.87
12.68
3.45
16.98
2.44
14.38
Route of Administration
(In the past month, please
report the percentage of marijuana you consumed
in each of the following ways (must total to 100%))
Smoked in joint/blunt without tobacco
32.68
36.57
31.25
35.62
22.21
32.79
29.49
35.62
Smoked in joint/blunt with tobacco
5.94
18.96
4.67
16.69
4.13
15.96
5.06
17.60
Smoked in bong/water pipe without tobacco
20.93
30.53
21.02
29.93
18.73
31.09
20.35
30.47
Smoked in bong/water pipe with tobacco
2.74
12.93
3.21
13.44
2.02
11.34
2.67
12.58
Smoked in bowl/pipe without tobacco
19.76
31.10
20.08
30.36
12.63
26.95
18.13
30.07
Smoked in bowl/pipe with tobacco
2.38
11.61
1.81
10.40
1.13
8.36
1.85
10.42
Eaten/cooked
10.73
24.87
10.83
25.88
15.26
30.42
11.88
26.77
Used in a vaporizer
5.79
18.67
8.07
21.61
23.90
35.83
11.11
26.03
Setting of Use
(In the past month, please report the
percentage of times that you used marijuana in
each of the following ways (must total to 100%))
At my home
32.88
38.49
38.09
39.99
53.18
40.97
40.11
40.58
At a friend’s home
32.94
37.12
35.11
37.16
24.54
33.98
31.36
36.55
At a stranger’s home
1.46
8.78
1.06
6.35
0.66
6.01
1.11
7.23
Outside
13.28
26.63
8.75
20.95
9.58
21.93
10.72
23.70
In a car
10.56
22.10
9.24
21.75
6.74
17.78
9.28
21.26
At a party
8.18
19.34
6.49
17.27
4.31
14.74
6.46
17.37
Other
1.46
10.55
0.36
3.55
0.98
8.79
1.23
9.64
Social Context of Use
(In the past month, please
report the percentage of times that you used
marijuana in each of the following ways (must total
to 100%))
Alone
15.88
27.01
17.40
28.11
30.42
36.00
20.03
30.44
With friends
76.28
33.09
75.53
33.34
58.84
40.08
71.73
35.77
With family
5.09
17.49
4.85
17.36
8.13
22.38
5.74
18.80
With people I don’t know [strangers]
1.68
8.51
1.29
7.13
0.75
5.80
1.30
7.39
Other [with others]
1.25
10.19
1.06
9.63
1.86
12.62
1.33
10.64
Unplanned Use
(In the past month, please report
the percentage of marijuana that you used in the
following way I did not make a plan to use
marijuana (0% to 100%))
44.60
42.31
38.00
41.50
40.17
41.77
40.94
41.96
Source of Cannabis
(In the past month, please
report the percentage ofmarijuana that you used
from the following sources (must total 100%):)
I bought it from a dispensary in the state where I
live [dispensary1]
6.56
22.05
14.13
31.56
24.07
40.09
13.92
31.80
I bought it from a dispensary in the state where I
do not live [dispensary2]
1.88
11.01
2.76
13.62
3.69
16.27
2.67
13.55
I bought it, but not from a dispensary [black
market]
33.51
41.32
29.90
39.50
23.98
37.87
29.82
39.96
Idid not buy it [Did not purchase]
58.84
43.83
53.77
44.60
48.26
45.83
54.12
44.80
Note. MOST = Marijuana Outcomes Study Team, PSST = Protective Strategies Study Team, ART = Addictions Research
Team
Cannabis, A Publication of the Research Society on Marijuana
Cannabis Protective Behavioral Strategies
(PBS).
We used the mean of the 17-item version
(Pedersen et al., 2017) of the PBSM (Pedersen et al.,
2016) to assess cannabis PBS use. Internal
consistency was high in each sample (α = .903, .925,
.902). The PBSM has been shown to be a robust
protective factor associated with lower cannabis use
(severity) and consequences (Pearson et al., 2017;
Pedersen et al., 2018).
Negative Cannabis-Related Consequences.
We
used the sum of the 21-item version of the Marijuana
Consequences Questionnaire (Simons et al., 2012) to
measure negative cannabis-related consequences.
Internal consistency was high in each sample (α =
.859, .886, .879).
Cannabis use severity.
We used the sum of the 8-
item Cannabis Use Disorder Identification Test—
Revised (CUDIT-R) (Adamson et al., 2010) to
measure CUD symptoms. Internal consistency was
adequate in each sample (α = .816, .833, .837).
Analysis Plan
We examined the context of use variables with
means and standard deviations across the three
datasets individually and joined as one dataset.
Additionally, we wanted to examine potential
predictive ability of contextual factors of use on
cannabis use outcomes (i.e., cannabis PBS, negative
cannabis-related consequences, and cannabis use
disorder severity). We used machine learning,
specifically random forests, to examine the potential
for contextual factors to predict outcomes. Random
forests are an extension of regression trees (Breiman,
2001). Regression trees use a nonparametric
algorithm to create a split, or a point in a predictor
that best separate the outcome variable (Strobl et al.,
2009). In traditional regression trees, the output
provides a single tree, or a visual representation of
the algorithm’s classification of the outcome. In
random forests, hundreds of trees are created by
randomly subsampling predictor variables at each
split, and then averaging the predictions of each tree
to find what variables are most important in
predicting the outcome (Breiman, 2001). The same
random forests procedures can also be used to impute
missing data (Tang & Ishwaran, 2017).
First, we used the
missForest
(Stekhoven, 2022)
package to impute all of the missing data via random
forest imputation. Then, we separated the data into
a training dataset (80% of the available data) that we
used to run the initial random forest model and a
testing dataset (20% of the available data) reserved
to test the predictive ability of the model. Splitting
the data in this way reduces the chances of the
algorithm finding random variance and overfitting
the model, as well as improves generalizability (Ho et
al., 2020). We used the
randomForest
package (Liaw
& Wiener, 2002) in
R
(R Core Team, 2023) to find the
optimal number of random predictors (i.e., tuning) for
the model to subsample at each split (mtry). Then, we
ran a random forest model for each outcome variable
(three models) with their respective tuning
parameters with the training dataset. Finally, we
used the random forest model to make predictions on
the testing dataset. We report the mean absolute
error (MAE; average distance between predicted and
actual values), the mean squared error (average
squared difference between predicted and actual
values), the root mean squared error (root squared
MSE), and the proportion of variance in the outcome
explained by the model (
R
2). Each model consisted of
only contextual factors as predictors. The MAE and
RMSE are dependent upon the scale (range) of the
outcome variable, and therefore there are no general
guidelines for what constitutes “acceptable” fit.
However, lower values of the MAE and RMSE
indicate a more accurate prediction.
RESULTS
The means and standard deviations of the
percentages of endorsement across all contexts of use
are reported in Table 1. In the results presented
below, we report noticeable trends in all three
datasets. We also include bivariate correlations
between all contextual indicators and the three
outcome variables to determine the directional
relationship between the contextual factors and
outcomes (Table 2).
Money Spent and Intoxication
Overall, participants reported they spent an
average of $46.19 on cannabis in the 30 days prior
to study participation. The amount of money
spent increased slightly between project MOST to
project PSST and again from project PSST to
project ART.
Cannabis Use Context
Table 2.
Raw correlations between cannabis use context variables and cannabis-related outcomes across each
dataset
MOST
PSST
ART
1
2
3
1
2
3
1
2
3
1. CUDIT-R
2. MACQ
.629**
.609**
.649**
3. PBSM
-.424**
-.363**
-.397**
-.291**
-.489**
-.363**
4. Source of Cannabis (dispensary1)
.108**
.086**
-.078**
.165**
.097**
-.068**
.162**
.098**
-.120**
5 Source of Cannabis (dispensary2)
.067
.035
-.055*
.014
.000
-.027
.037
.046
-.055*
6. Source of Cannabis (Black market)
.355**
.292**
-.318**
.362**
.257**
-.287**
.266**
.198**
-.236**
7 Source of Cannabis (Did not
purchase)
-.401**
-.328**
.344**
-.437**
-.290**
.306**
-.376**
-.266**
.319**
8. Money Spenta
.483**
.408**
-.448**
.526**
.352**
-.405**
.505**
.365**
-.428**
9. Typical Intoxa
.217**
.155**
-.191**
.255**
.137**
-.175**
.182**
.130**
-.143**
10. Peak Intoxa
.322**
.218**
-.255**
.364**
.223**
-.251**
.310**
.224**
-.227**
11. Peak Frequencya
.049
.010
-.069**
.067**
.001
-.070**
-.056*
-.053*
.021
12. Length of Intoxa
.008
.003
-.044*
.072**
.026
-.056*
.028
.046
-.013
13. Form of Cannabis (Plant)
.050
.010
.016
.071**
.055*
-.013
.070*
.023
-.079**
14. Form of Cannabis (Edibles)
-.110**
-.085**
.099**
-.167**
-.134**
.160**
-.211**
-.154**
.148**
15. Form of Cannabis (Concentrates)
.041
.080**
-.132**
.133**
.088**
-.149**
.139**
.123**
-.026
16. Form of Cannabis (Other form)
.016
-.004
.006
-.102**
-.037
-.011
-.084**
-.037
-.023
17. Route of Administration (joint)
-.039
-.009
.003
-.057*
-.026
-.006
.004
-.055*
.012
18. Route of Administration (joint
tobacco)
.094*
.067**
-.025
.053*
.049*
-.034
.045
-.013
-.044
19. Route of Administration (bong)
.134**
.073**
-.090**
.159**
.129**
-.096**
.167**
.154**
-.141**
20. Route of Administration (bong
tobacco)
.107**
.084**
-.039
.119**
.112**
-.090**
.109**
.103**
-.097**
21. Route of Administration (bowl)
-.065
-.054*
.054*
.000
-.048*
.034
-.040
-.014
.027
22. Route of Administration (bowl
tobacco)
-.018
-.007
-.008
-.005
.025
-.023
-.016
.030
-.002
23. Route of Administration (eaten)
-.161**
-.072**
.072**
-.181**
-.136**
.175**
-.210**
-.162**
.128**
24. Route of Administration
(vaporizer)
.035
-.036
.002
-.015
-.021
-.023
.007
.030
.034
25. Setting of Use (At home)
.211**
.141**
-.152**
.199**
.123**
-.121**
.206**
.140**
-.184**
26. Setting of Use (At friend's)
-.163**
-.101**
.155**
-.190**
-.119**
.138**
-.218**
-.162**
.177**
27. Setting of Use (At stranger's)
-.035
.010
-.010
.006
.006
-.087**
-.006
.021
.002
28. Setting of Use (outside)
-.023
-.036
-.010
-.014
-.007
-.025
-.059*
-.034
.064*
29. Setting of Use (car)
.012
.042
-.030
.014
.030
.009
.101**
.054*
-.068*
30. Setting of Use (party)
-.090*
-.072**
.055*
-.049*
-.070**
.046*
-.051
-.024
.070**
31. Setting of Use (Other place)
.014
-.028
-.009
-.023
.005
-.010
-.091**
-.025
.030
32. Social Context of Use (Alone)
.300**
.238**
-.261**
.332**
.224**
-.272**
.278**
.185**
-.297**
33. Social Context of Use (With
friends)
-.304**
-.217**
.265**
-.281**
-.182**
.257**
-.249**
-.166**
.287**
34. Social Context of Use (With
family)
.022
-.005
-.056*
-.015
-.035
-.017
.032
.016
-.007
35. Social Context of Use (With
strangers)
.074
.067**
-.042
.046*
.080**
-.091**
-.025
.002
-.014
36. Social Context of Use (With
others)
.058
.019
-.039
-.009
-.002
-.005
-.048
-.030
-.044
37. Unplanned Usea
-.348**
-.200**
.206**
-.280**
-.141**
.125**
-.175**
-.105**
.067*
Note
. *
p
< .05, **
p
< .01, a Proxy of direct use or not a social context and removed from analyses.
Cannabis, A Publication of the Research Society on Marijuana
Participants were asked to rate on a 0 – 100
scale how high they typically get when they use
cannabis. In each of the three studies,
participants reported percentages in the 60-65
range with an overall average of 61.835.
Participants were also asked to rate their
highest level of intoxication on the same scale.
In all three studies, participants reported the
highest level of intoxication in the 70s (Overall
M
= 73.151) and that they achieve this peak
intoxication over half of the times they use
cannabis (Overall
M
= 59.487%). On days
participants used cannabis in the past month
they reported feeling high for an average of
3.392 hours.
Cannabis Form
In all three studies, participants reported
using cannabis flower most of the time (Overall
M
= 70.384%). However, a notable drop in
cannabis flower use occurred between project
MOST (
M
= 78.646%) and project ART (
M
=
53.265%). The drop in the use of flower
corresponded with a similar increase in the use
of concentrates (MOST
M
= 8.621%; ART
M
=
26.073%).
Route of Administration
The most prominent route of administration
in projects MOST and PSST was smoking a joint
or blunt without tobacco. Participants in project
ART endorsed using a joint or blunt without
tobacco (
M
= 22.205%) and using a vaporizer (
M
= 23.902%) at similar rates. Reported vaporizer
use in project ART was much higher than
reported vaporizer use in projects MOST (
M
=
5.787%) and PSST (
M
= 8.072%).
Use Settings
Overall, using cannabis at home was the
most endorsed setting. However, in project
MOST and PSST, participants tended to use at
home or at a friend’s house at about the same
frequencies. Compared to MOST and PSST,
participants in project ART appeared to make a
trade-off between using cannabis at their own
house (
M
= 53.180%) and their friend’s house
(24.540%).
Social Context of Use
Overall, participants reported mostly using
cannabis with their friends in all three studies.
One notable difference between the three studies
is that participants in project ART reported using
alone (
M
= 30.418%) more often than participants
in MOST (
M
= 15.884%) and PSST (
M
= 17.396%).
Source of Cannabis and Money Spent
Across all three projects, participants mostly
endorsed not sourcing cannabis themselves. In
projects MOST and PSST, the second most
endorsed source was sourcing cannabis from a
place other than a dispensary. In project ART, the
second most endorsed source was obtaining
cannabis from a dispensary. In all studies, below
5% of cannabis sourcing involved crossing state
lines to purchase at a dispensary in another state.
Random Forest Models
First, we combined all three datasets and
imputed missing values using the
missForest
package (Stekhoven, 2022). The number of
missing values for the MACQ (MOST = 1.97%;
PSST = 1.89%, ART = 1.58%) and PBSM (MOST
= 2.27%; PSST = 1.75%, ART = 1.58%) was
acceptable in all three datasets. Regarding the
CUDIT-R, participants in project MOST were
randomly assigned to complete one of four
measures of cannabis use disorder symptoms, one
of which was the CUDIT-R. Thus, missingness for
the CUDIT-R in project MOST was high (69.57%).
Missingness for the CUDIT-R in projects PSST
and ART were acceptable (PSST = 1.80%, ART =
4.94%). The
missForest
package subsets the data
into complete cases and variables with missing
data. The package then runs a random forest
algorithm based on the observed values to impute
a value for missing data (Stekhoven, 2022).
After data imputation, we split the dataset
into a training dataset and a testing dataset. For
each outcome (PBSM, MACQ, CUDIT-R) we
conducted a tuning model that examined the
optimal number of variables randomly sampled at
each split of the decision trees. The optimal number
of variables randomly sampled for the CUDIT-R,
PBSM, and MACQ models was 5. Finally, we
conducted random forest models for all three
variables using the selected number of splits, 500
Cannabis Use Context
decision trees, in the
randomForest
package (Liaw
& Wiener, 2002) in
R
(R Core Team, 2022). Below,
we report the variable importance, or predictive
utility of a variable across all the decision trees in a
random forest, from the training dataset and model
fit from using the training datasets on the testing
datasets.
For the PBSM, our rank-ordered variable
importance plot can be viewed in Figure 1. Using
alone, using with friends, obtaining cannabis on the
black market, using concentrate, and using at home
were the most important variables in predicting the
PBSM. However, when using the random forest
model to predict values in the testing dataset, the
random forest predictions had room for
improvement (Table 3). On average, our model’s
predicted values deviated from the true values
(MAE) by 0.74 units of the PBSM (range 1 – 6). The
squared differences between the predicted and
actual values (MSE) was 0.94, and our model
accounted for 16% of the variance in PBSM scores.
For the MACQ, our rank ordered variable
importance plot can be viewed in Figure 2. Using
alone, obtaining cannabis on the black market,
using with friends, primarily using a bong, and
using at home were the most important predictors
for the MACQ. We used the training model to
predict MACQ scores in the portion of data set aside
for predictions (Table 3). On average, our model’s
predicted values deviated from the true values
(MAE) by 2.68 units of the MACQ (range 0 – 21).
The squared differences between the predicted and
actual values (MSE) was 13.02. Overall, our random
forest model of contextual factors accounted for 17%
of the variance in the MACQ.
The rank ordered variable importance plot for
our random forest model predicting the CUDIT-R
can be viewed in Figure 3. The most important
variables in predicting the CUDIT-R were using
alone, obtaining cannabis on the black market,
using with friends, using at home, and using a bong.
On average, the model’s predicted values deviated
from the true values (MAE) by 3.12 units of the
CUDIT-R sum (range = 0 – 32). The squared
differences between the predicted and actual values
was 18.79 and the model accounted for 38% of the
variance in the CUDIT-R sum.
Figure 1.
Plot of variable importance for the PBSM in order from least important (top) to most
important (bottom).
Note
. The (+) and (-) after each contextual variable indicates the directional relationship to the PBSM.
Cannabis, A Publication of the Research Society on Marijuana
Figure 2.
Plot of variable importance for the MACQ in order from least important (top) to most
important (bottom).
Note
. The (+) and (-) after each contextual variable indicates the directional relationship to the MACQ.
Figure 3.
Plot of variable importance for the CUDIT-R in order from least important (top) to most
important (bottom).
Note
. The (+) and (-) after each contextual variable indicates the directional relationship to the CUDIT-R.
Cannabis Use Context
Table 3.
Random Forest Fit Statistics
Outcome
mtry
R2
MAE
MSE
RMSE
Protective Behavioral
Strategies (PBSM)
5
0.16
0.75
0.94
0.97
Negative Consequences
(MACQ)
5
0.17
2.68
13.02
3.46
Cannabis Use Severity
(CUDIT-R)
5
0.46
2.89
16.59
4.07
Note
. PBSM= Protective Behavioral Strategies for Marijuana, MACQ = Marijuana Consequences
Questionnaire, CUDIT-R = Cannabis Use Disorder Identification Test-Revised, mtry= the optimal
number of random predictors (i.e., tuning) for the model to subsample at each split, MAE = Mean
Absolute Error, MSE = Mean Squared Error, RMSE = Root Mean Squared Error.
DISCUSSION
Overall, the present study extends the
previous literature, which has largely focused on
using a few contextual variables to determine
factor structures or latent profiles (Beck et al.,
2009; Gray et al., 2024; Spinella et al., 2019), by
providing descriptive statistics across a broad
array of social contexts of cannabis use. Regarding
the form of cannabis used, college students appear
to predominately use flower cannabis, though the
use of edibles and concentrates was not minimal.
Bivariate correlations between cannabis form and
cannabis outcomes (Table 2) indicated that edible
usage was most consistently correlated (compared
to other forms of cannabis) with less use disorder
severity and consequences, and more PBS use.
Though, the use of concentrates was significantly
correlated with less PBS use. Of note regarding
cannabis form, the use of concentrate was the
most important cannabis form predictor in all
three random forest models. This in part could be
due to the greater exposure to THC when using
concentrates versus flower (Bidwell et al., 2020).
Regarding different routes of administration,
the use of a bong was most consistently correlated
with increased consequences and disorder
severity and decreased use of PBS, and similar to
form results, eating cannabis appeared to be the
most protective route of administration (Table 2).
Additionally, using a bong tended to be the most
important route of administration in the random
forest models, other than for PBS, where using a
joint was slightly more important. Bongs tend to
be relatively indiscreet and would likely be owned
by individuals that consume cannabis regularly,
though more work is needed to determine why
bong use specifically may be associated with
worse outcomes. College students predominately
consume using joints without tobacco and co-use
with tobacco was not highly endorsed.
Regarding direct social contexts of use (who
participants used with), participants tended to
use with friends, and using with friends was the
most consistent social context correlated with
fewer consequences and disorder severity and
more PBS use. In contrast, using alone was the
most consistent social context correlated with
negative cannabis outcomes, consistent with
previous literature (Table 2; Buckner et al., 2016;
Creswell et al., 2015). Using alone was the most
important contextual factor in all of the random
forest models. Considering that solitary use
accounts for much of the relationship between
social anxiety and poor cannabis use outcomes
(Buckner et al ., 2016), it may be that solitary use
is more associated with negative reinforcement, or
using to remove unwanted emotional states.
Given the fact that using alone was the most
important predictor in all models, this may
highlight the need for preventative and clinical
treatments to focus on decreasing the amount of
time individuals use cannabis alone. Regarding
where individuals used, using at a friend’s house
was consistently correlated with positive cannabis
outcomes, while using at home was associated
with negative outcomes (Table 2). Additionally,
using cannabis at home was the most important
social setting in all three random forest models.
Using at home and using alone are potentially
conflated and our models cannot differentiate
whether participants used at home alone or with
friends. Future work should focus on examining
social networks of individuals that often use at
home and whether including others may have
protective effects on cannabis outcomes.
Cannabis, A Publication of the Research Society on Marijuana
Regarding source of cannabis purchase, the most
protective source of purchase was not purchasing
cannabis, and purchasing on the black market
was the source most strongly correlated with
negative outcomes (Table 2). In fact, sourcing
cannabis on the black market was the most
important source context in all of the random
forest models. This likely indicates individuals
that often purchase cannabis or go out of their
way to source cannabis in places that do not have
the same tax burden as a legalized market. One
important note is that crossing state lines to
source cannabis was not highly endorsed in any of
the studies.
Lastly, the predictive models accounted for
varying proportions of the cannabis outcomes’
variances. Specifically, contextual factors
accounted for 16% and 17% of the variance in PBS
and cannabis use consequences respectively and
38% of variance in the CUDIT-R. Considering our
models removed proxies of use (i.e., level of
intoxication, money spent), our results indicate
that contextual factors likely account for
additional variance in cannabis-related outcomes
beyond direct use. Additionally, the random forest
models were able to predict outcomes relatively
well. Recalling that the MAE is the average error
of the model’s prediction in the same scale as the
outcome, the model's relative errors were all
within 9%-13% of the outcome variables’ range
and may provide a benchmark for future studies
using machine learning with cannabis contexts.
Limitations
Our study has several limitations. First, we
created our contextual measurement tool, and
said tool has not been validated for real-time use.
Second, our models do not account for the legal
status of cannabis in the participant’s state of
residence. As such, we do not know the status of
how participants sourced cannabis. While the
rates of crossing state lines to obtain cannabis
were low in all three studies, the rates may
change depending on the legal status of each
state, and how far away the participants were
from a dispensary. Lastly, while the studies were
conducted over a span of 4 years, we do not make
any assumptions regarding the trends of cannabis
use as state-level legalization has become more
widespread over time. This is especially relevant
as the COVID-19 pandemic had not occurred
during PSST and MOST data collection but had
already occurred during the ART data collection.
Future work should focus on changes in these
contextual trends and determine how potential
changes may affect outcomes.
Future Directions
Much work needs to be done regarding the
contexts surrounding cannabis use. Specifically,
participants in our study tended to report using
cannabis with friends. Under the assumption that
friends using cannabis together are not using
their own pipes, bongs, or joints, it is likely
difficult to accurately measure the amount of
cannabis consumed by everyone, even if the
weight and potency are known prior to group
consumption. For example, even if study
participants are asked to report the potency and
to pre-measure the weight of each joint/bowl in
real time, there is no way to know what
percentage of that weight in combusted THC that
everyone in a group session is consuming. This
predicament contrasts with alcohol use, where
more accurate measurements can be assumed by
standard drink conversions. It may be that
measuring additional contexts such as subjective
intoxication, money spent, and form of cannabis
use can be appropriate proxies for precise dosage
and weights.
The incorporation of assessing contexts of use
could also provide pertinent information
regarding environmental factors related to use.
Implementing contextual measures from a
theoretical framework could help improve
existing models predicting cannabis use
outcomes. That is, what are the effects of core
predictors of cannabis use outcomes when
incorporating environmental factors into existing
models? Much of the modeling on cannabis use
outcomes examine outcomes as functions of use,
emotions, or urges. However, it is likely the
predictors fluctuate between different contexts of
use.
REFERENCES
Adamson, S. J., Kay-Lambkin, F. J., Baker, A. L.,
Lewin, T. J., Thornton, L., Kelly, B. J., &
Sellman, J. D. (2010). An improved brief
measure of cannabis misuse: The Cannabis
Use Disorders Identification Test-Revised
Cannabis Use Context
(CUDIT-R).
Drug and Alcohol Dependence
,
110
(1–2), 137–143.
https://doi.org/10.1016/j.drugalcdep.2010.02.0
17
Bidwell, L. C., YorkWilliams, S. L., Mueller, R. L.,
Bryan, A. D., & Hutchison, K. E. (2018).
Exploring cannabis concentrates on the legal
market: User profiles, product strength, and
health-related outcomes.
Addictive Behaviors
Reports
,
8
(March), 102–106.
https://doi.org/10.1016/j.abrep.2018.08.004
Bravo, A. J., Pearson, M. R., Conner, B. T., &
Parnes, J. E. (2017). Is 4/20 an event-specific
marijuana holiday? A daily diary investigation
of marijuana use and consequences among
college students.
Journal of Studies on Alcohol
and Drugs
,
78
(1), 134–139.
https://doi.org/10.15288/jsad.2017.78.134
Bravo, A. J., Villarosa-Hurlocker, M. C., &
Pearson, M. R. (2018). College Student Mental
Health: An Evaluation of the DSM-5 Self-
Rated Level 1 Cross-Cutting Symptom
Measure.
Psychological Assessment
,
30
(10),
1382–1389.
https://doi.org/10.1037/pas0000628
Breiman, L. (2001). Random Forests.
Machine
Learning
,
45
(1), 5–32.
https://doi.org/10.1023/A:1010933404324
Buckner, J. D., Ecker, A. H., & Dean, K. E. (2016).
Solitary cannabis use frequency mediates the
relationship between social anxiety and
cannabis use and related problems.
American
Journal on Addictions
,
25
(2), 99–104.
https://doi.org/10.1111/ajad.12339
Cloutier, R. M., Calhoun, B. H., & Linden-
Carmichae, A. N. (2022). Associations of mode
of administration on cannabis consumption
and subjective intoxication in daily life.
Psychology of Addictive Behaviors
,
36
(1), 67–
77. https://doi.org/10.1037/adb0000726
Creswell, K. G., Chung, T., Clark, D. B., & Martin,
C. S. (2015). Solitary cannabis use in
adolescence as a correlate and predictor of
cannabis problems.
Drug and Alcohol
Dependence
,
156
(2015), 120–125.
https://doi.org/10.1016/j.drugalcdep.2015.08.0
27
D’Amico, E. J., Rodriguez, A., Dunbar, M. S.,
Firth, C. L., Tucker, J. S., Seelam, R.,
Pedersen, E. R., & Davis, J. P. (2020). Sources
of cannabis among young adults and
associations with cannabis-related outcomes.
International Journal of Drug Policy
,
86
,
102971.
https://doi.org/10.1016/j.drugpo.2020.102971
Gray, K. M., Watson, N. L., & Christie, D. K.
(2009). Challenges in quantifying marijuana
use.
American Journal on Addictions
,
18
(2),
178–179.
https://doi.org/10.1080/10550490902772579
Ho, S. Y., Phua, K., Wong, L., & Bin Goh, W. W.
(2020). Extensions of the external validation
for checking learned model interpretability
and generalizability.
Patterns
,
1
(8), 100129.
https://doi.org/10.1016/j.patter.2020.100129
Lemberger, L., Martz, R., & Rodda, B. (1973).
Comparative pharmacology of Δ9
tetrahydrocannabinol and its metabolite, 11
OH Δ9 tetrahydrocannabinol.
Journal of
Clinical Investigation
,
52
(10), 2411–2417.
https://doi.org/10.1172/JCI107431
Liaw, A., & Wiener, M. (2002). Classification and
regression by randomForest.
R News, 2
(3), 18–
22. https://cran.r-
project.org/web/packages/randomForest/
Parnes, J. E., Bravo, A. J., Conner, B. T., &
Pearson, M. R. (2018). A burning problem:
Cannabis lessons learned from Colorado.
Addiction Research and Theory
,
26
(1), 3–10.
https://doi.org/10.1080/16066359.2017.131541
0
Pearson, M. R., Bravo, A. J., & Protective
Strategies Study Team. (2019). Marijuana
protective behavioral strategies and
marijuana refusal self-efficacy: Independent
and interactive effects on marijuana-related
outcomes.
Psychology of Addictive Behaviors
,
33
(4), 412–419.
https://doi.org/10.1037/adb0000445
Pearson, M. R., & Henson, J. M. (2013).
Unplanned drinking and alcohol-related
problems: A preliminary test of the model of
unplanned drinking behavior.
Psychology of
Addictive Behaviors
,
27
(3), 584–595.
https://doi.org/10.1037/a0030901
Pearson, M. R., Liese, B. S., Dvorak, R. D.,
Anthenien, A. M., Bravo, A. J., Conner, B. T.,
Correia, C. J., Dvorak, R. D., Egerton, G. A.,
Hustad, J. T. P., Kholodkov, T., King, K. M.,
Liese, B. S., Messina, B. G., Murphy, J. G.,
Neighbors, C., Nguyen, X. T., Parnes, J. E.,
Pearson, M. R., … Read, J. P. (2017). College
student marijuana involvement: Perceptions,
use, and consequences across 11 college
Cannabis, A Publication of the Research Society on Marijuana
campuses.
Addictive Behaviors
,
66
, 83–89.
https://doi.org/10.1016/j.addbeh.2016.10.019
Pedersen, E. R., Huang, W., Dvorak, R. D., Prince,
M. A., Hummer, J. F., & (The Marijuana
Outcomes Study Team) (2017). The Protective
Behavioral Strategies for Marijuana Scale:
Further examination using item response
theory.
Psychology of Addictive Behaviors
,
31
(5), 548–559.
https://doi.org/10.1037/adb0000271
Pedersen, E. R., Hummer, J. F., Rinker, D. V.,
Traylor, Z. K., & Neighbors, C. Measuring
Protective Behavioral Strategies for
Marijuana Use Among Young Adults (2016).
Journal of Studies on Alcohol and Drugs,
77
(3), 441-50.
https://doi.org/10.15288/jsad.2016.77.441
Pedersen, E. R., Villarosa-Hurlocker, M. C., &
Prince, M. A. (2018). Use of Protective
Behavioral Strategies among Young Adult
Veteran Marijuana Users.
Cannabis
,
1
(1), 14–
27.
https://doi.org/10.26828/cannabis.2018.01.002
Prince, M. A., Conner, B. T., & Pearson, M. R.
(2018). Quantifying Cannabis: A field study of
marijuana quantity estimation.
Psychology of
Addictive Behaviors
,
32
(4), 426–433.
https://doi.org/10.1037/adb0000370
R Core Team. (2019). R: A language and
environment for statistical computing. R
Foundation for Statistical Computing, Vienna,
Austria. https://www.R-Project.org/
Richards, D. K., Pearson, M. R., & Field, C. A.
(2021). A Comprehensive Examination of
Alcohol-Related Motivations Among College
Students: Unique Relations of Drinking
Motives and Motivations for Drinking
Responsibly.
Experimental and Clinical
Psychopharmacology
.
https://doi.org/10.1037/pha0000526
Richards, D. K., Schwebel, F. J., Field, C. A., &
Pearson, M. R. (2023). The Associations of
Basic Psychological Need Satisfaction and
Need Frustration with Cannabis-Related
Outcomes in a Multi-Site Sample of College
Students.
Journal of Psychoactive Drugs
,
00
(00), 1–10.
https://doi.org/10.1080/02791072.2023.219160
5
Richards, D. K., Schwebel, F. J., Sotelo, M.,
Pearson, M. R., & Marijuana Outcomes Study
Team. (2021). Self-Reported Symptoms of
Cannabis Use Disorder: Psychometric testing
and validation.
Experimental and Clinical
Psychopharmacology
,
29
(2), 157–165.
https://doi.org/10.1037/pha0000455
Schubart, C. D., Boks, M. P. M., Breetvelt, E. J.,
van Gastel, W. A., Groenwold, R. H. H.,
Ophoff, R. A., Sommer, I. E. C., & Kahn, R. S.
(2011). Association between cannabis and
psychiatric hospitalization.
Acta Psychiatrica
Scandinavica
,
123
(5), 368–375.
https://doi.org/10.1111/j.1600-
0447.2010.01640.x
Schwebel, F. J., Richards, D. K., Pfund, R. A.,
Joseph, V. W., & Pearson, M. R. (2022). Using
Decision Trees to Identify Salient Predictors of
Cannabis-Related Outcomes.
Journal of
Psychoactive Drugs
, 1–10.
https://doi.org/10.1080/02791072.2021.201408
1
Schwilke, E. W., Schwope, D. M., Karschner, E. L.,
Lowe, R. H., Darwin, W. D., Kelly, D. L.,
Goodwin, R. S., Gorelick, D. A., & Huestis, M.
A. (2009). Δ9-tetrahydrocannabinol (THC), 11-
hydroxy-THC, and 11-nor-9-carboxy-THC
plasma pharmacokinetics during and after
continuous high-dose oral THC.
Clinical
Chemistry
,
55
(12), 2180–2189.
https://doi.org/10.1373/clinchem.2008.122119
Simons, J. S., Dvorak, R. D., Merrill, J. E., &
Read, J. P. (2012). Dimensions and severity of
marijuana consequences: Development and
validation of the Marijuana Consequences
Questionnaire (MACQ).
Addictive Behaviors
,
37
(5), 613–621.
https://doi.org/10.1016/j.addbeh.2012.01.008
Spinella, T. C., Stewart, S. H., & Barrett, S. P.
(2019). Context matters: Characteristics of
solitary versus social cannabis use.
Drug and
Alcohol Review
,
38
(3), 316–320.
https://doi.org/10.1111/dar.12912
Stekhoven, D. J. (2022).
MissForest:
Nonparametric Missing Value Imputation
using Random Forest. R package version 1.5.
https://cran.r-
project.org/web/packages/missForest/index.ht
ml
Strobl, C., Malley, J., & Tutz, G. (2009). An
introduction to recursive partitioning:
Rationale, application, and characteristics of
classification and regression trees, bagging,
and random Forests.
Psychological Methods
,
Cannabis Use Context
14
(4), 323–348.
https://doi.org/10.1037/a0016973
Tang, F., & Ishwaran, H. (2017). Random forest
missing data algorithms.
Statistical Analysis
and Data Mining: The ASA Data Science
Journal
,
10
(6), 363–377.
https://doi.org/10.1002/sam.11348
Funding and Acknowledgements:
Matison
McCool was supported in part by the National
Institute on Alcohol Abuse and Alcoholism,
T32AA018108 (PI: Witkiewitz).
Conflict of interest statement:
We have no
conflict of interest to declare
.
We would like to acknowledge the efforts of
Sarah L. Simons with conducting literature
searches and contributing to an early version of
this manuscript. Data were collected by three
research teams: Marijuana Outcomes Study
Team (MOST), Protective Strategies Study Team
(PSST), and the Addictions Research Team
(ART).
*MOST includes the following investigators (in
alphabetical order): Amber M. Anthenien,
University of Houston; Adrian J. Bravo,
University of New Mexico; Bradley T. Conner,
Colorado State University; Christopher J.
Correia, Auburn University; Robert D. Dvorak,
University of Central Florida; Gregory A.
Egerton, University at Buffalo; John T. P.
Hustad, Pennsylvania State University College
of Medicine; Tatyana Kholodkov, University of
Wyoming; Kevin M. King, University of
Washington; Bruce S. Liese, University of
Kansas; Bryan G. Messina, Auburn University;
James G. Murphy, The University of Memphis;
Clayton Neighbors, University of Houston; Xuan-
Thanh Nguyen, University of California, Los
Angeles; Jamie E. Parnes, Colorado State
University; Matthew R. Pearson, University of
New Mexico; Eric R. Pedersen, RAND; Mark A.
Prince, Colorado State University; Sharon A.
Radomski, University at Buffalo; Lara A. Ray,
University of California, Los Angeles; Jennifer P.
Read, University at Buffalo.
**PSST includes the following investigators:
Matthew R. Pearson, University of New Mexico
(Coordinating PI); Adrian J. Bravo, University of
New Mexico (Co-PI); Mark A. Prince, Colorado
State University (site PI); Michael B. Madson,
University of Southern Mississippi (site PI);
James M. Henson, Old Dominion University (site
PI); Alison Looby, University of Wyoming (site
PI); Vivian M. Gonzalez, University of Alaska-
Anchorage (site PI); Amber M. Henslee, Missouri
Science & Technology (site PI); Carrie Cuttler,
Washington State University (site PI), Maria M.
Wong, Idaho State University (site PI), Dennis E.
McChargue, University of Nebraska-Lincoln (site
PI).
***ART includes the following investigators:
Matthew R. Pearson, University of New Mexico
(Coordinating PI); Adrian J. Bravo, William &
Mary (site PI); Bradley T. Conner, Colorado
State University – Fort Collins (site PI); Carrie
Cuttler, Washington State University (site PI);
Craig A. Field, University of Texas at El Paso
(site PI); Vivian Gonzalez, University of Alaska -
Anchorage (site PI); James M. Henson, Old
Dominion University (site PI); Jon M. Houck,
Mind Research Network; Kevin M. King,
University of Washington (site PI); Benjamin O.
Ladd, Washington State University (site PI);
Kevin S. Montes, California State University –
Dominguez Hills (site PI); Mark A. Prince,
Colorado State University – Fort Collins (site
PI); Maria M. Wong, Idaho State University (site
PI).
Copyright: © 2024 Authors et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction, provided the original author and
source are credited, the original sources is not
modified, and the source is not used for
commercial purposes.
