Predicting Heavy Alcohol Use Among Adolescents

American Journal of Orthopsychiatry
Copyright 2005 by the Educational Publishing Foundation
2005, Vol. 75, No. 1, 102–116
0002-9432/05/$12.00
DOI: 10.1037/0002-9432.75.1.102
Predicting Heavy Alcohol Use Among Adolescents
J. Greg Getz, PhD
James H. Bray, PhD
University of Houston—Downtown and Baylor
Baylor College of Medicine
College of Medicine
A prospective, school-based study of alcohol use in a multiethnic sample of 3,675 adolescents
found that family factors, other drug use, psychosocial– behavioral factors, and ethnic status are
important discriminators of adolescents who are heavy alcohol users when compared with those
who are experimental or moderate users. Implications for prevention and intervention are dis-
cussed.
For several decades, researchers have explored
1980; Windle, 1996). Various typologies or stage-
both the prevalence and the etiology of adolescent
process models of alcohol involvement have been
alcohol use, concluding that prevalence is high
used by researchers without the emergence of a
among middle and high school students in the
strong consensus that one particular model has
United States (Johnston, O’Malley, & Bachman,
greatest utility for all research purposes. Knupfer
1991, 1994, 2001) and that alcohol use is regarded
(1989) suggested an eight-stage model ranging
by many researchers to be statistically normative
from lifelong abstainer to frequent drunk. Werch
(Windle, 1999). Apart from documenting preva-
and DiClemente (1994) offered a ?ve-stage moti-
lence and demographic variability, early research
vational model ranging from precontemplation (se-
focused on risk factor analysis to identify cross-
riously thinking about initiating) to maintenance
sectional, multivariate correlates of adolescent al-
(continuing usage). Windle (1996) advocated a
cohol use and other problem behaviors (Merrill,
?ve-stage model ranging from abstainers to prob-
Kleber, Schwartz, Liu, & Lewis, 1999). Also, ef-
lem drinkers. Many adolescents use alcohol exper-
forts have been directed toward the testing of mod-
imentally, sometimes frequently and sometimes
els that hypothesize proximal and distal effects
consuming multiple drinks per occasion, without
involving mediated and moderated causal paths
engaging in other problem behaviors or experienc-
that connect risk and protective factors to alcohol
ing immediate negative consequences. Windle
use (Baer & Bray, 1999; Bray, Getz, & Baer, 2000;
(1996) suggested it is theoretically and empirically
Donovan, Jessor, & Costa, 1991; Huba & Bentler,
important to distinguish between heavy drinkers
1982; Zucker, Fitzgerald, & Moses, 1995).
and those who drink heavily and have experienced
Another important conceptual focus has been the
personal or social problems associated with drink-
qualitative distinction among stages of alcohol use,
ing. Such distinctions have important implications
such as initiation, experimentation, regular use,
for both understanding the multivariate etiology of
heavy use, and binge drinking (Kandel, 1975,
adolescent alcohol involvement and the design of
prevention/intervention programs.
Much research has been either purposely or in-
J. Greg Getz, PhD, Department of Social Sciences,
advertently directed toward the earlier stages of
University of Houston—Downtown, and Department of
initiation and experimentation (Reifman, Barnes,
Family and Community Medicine, Baylor College of Med-
icine; James H. Bray, PhD, Department of Family and
Dintcheff, Farrell, & Uhteg, 1998). Unintentional
Community Medicine, Baylor College of Medicine.
emphasis on initiation can arise because alcohol
A draft of this article was presented at the 109th Annual
use distributions are usually skewed toward the
Convention of the American Psychological Association,
low end of the scale, especially among samples of
San Francisco, August 2001.
young adolescents. In addition, use of regression
For reprints and correspondence: J. Greg Getz, PhD,
analytic techniques with continuous dependent
Department of Social Sciences, University of Houston—
variables can obscure the distinction between qual-
Downtown, 1 Main Street, Houston, TX 77002-1001, or
itatively different developmental stages of use or
James H. Bray, PhD, Department of Family and Commu-
abuse of alcohol. For these reasons, more attention
nity Medicine, Baylor College of Medicine, 3701 Kirby
Drive,
6th
Floor,
Houston,
TX
77098.
E-mail:
should be given to identifying predictors of heavy
getzg@uhd.edu or jbray@bcm.tmc.edu
use and problem-related use of alcohol among ad-
102

---

HEAVY ALCOHOL USE AMONG ADOLESCENTS
103
olescents (Knupfer, 1989; Windle, 1999). This
children or to monitor adequately their children’s
shift in focus is warranted because of the greater
activities. Under such circumstances, adolescents
deleterious health, behavioral, and social conse-
are more likely to experience stress, to engage in
quences of heavy or abusive adolescent alcohol
avoidance coping by using alcohol or other sub-
use, as opposed to experimentation, and because of
stances, and to associate with alcohol-using peers
the differential implications for the design of pre-
(Baer & Bray, 1999). Second, higher alcohol use
vention and intervention programs.
by mothers re?ects the potential for their children
to role model a dysfunctional coping strategy when
Psychosocial, Developmental Model of
family con?ict or stress occurs and to have alcohol
Adolescent Alcohol Use
more readily available for their own use. Alterna-
tively, an adolescent whose mother monitors his or
Previous research has demonstrated the utility of
her behavior and has open and high-quality com-
conceptualizing elements of the individuation pro-
munication with him or her is less likely to use
cess, stress, and peer alcohol usage as potential
alcohol (Bray et al., 2001; Dishion & McMahon,
mediators of the relationship between risk or pro-
1998; Duncan et al., 1995). It is unclear whether
tective family processes and adolescent alcohol
these variables discriminate both heavy use from
usage (Baer & Bray, 1999; Bell, Forthun, & Sun,
experimental/moderate use and nonuse from exper-
2000). These ?ndings are consistent among ado-
imental/moderate use.
lescents across gender and racial/ethnic back-
Psychosocial– behavioral constructs of stress,
ground, that is, non-Hispanic White, African
peer alcohol use, and deviant behavior are stable,
American, and Mexican American (Bray, Getz, &
positive predictors of adolescent alcohol use and
Baer, 2000). The possibility that these constructs
partially mediate the impact of family factors on
are mediators in the prediction of heavy alcohol
adolescent drinking (see Figure 1). Such mediation
use is the focus of this article. Distal, family pro-
has been demonstrated for stress and for peer use
cess-related constructs that are known to be pre-
of alcohol (Baer & Bray, 1999; Bray et al., 2000).
dictors of adolescent alcohol usage include family
Deviant behavior (a strong, positive correlate of
con?ict, parental alcohol use, quality of communi-
stress, peer alcohol use, and individual adoles-
cation with parents, and parental monitoring (Bray,
cents’ use) should function similarly. Again, it is
Adams, Getz, & Baer, 2001; Dishion & McMahon,
expected that these variables operate as risk fac-
1998; Duncan, Duncan, & Hops, 1996; Duncan,
tors, but their discriminatory strength is not clear
Tildesley, Duncan, & Hops, 1995). However, these
with regard to the distinction between heavy versus
constructs impact adolescent alcohol use, in part,
experimental/moderate use and nonuse versus ex-
indirectly through their effect on the individuation
perimental/moderate use.
process, use of other substances, expression of
Figure 1 represents the hypothesized set of re-
other deviant behavior, stress, and peer alcohol
lationships wherein the impact of various family
use, which, in turn, directly impact heavy alcohol
factors on heavy alcohol use is partially mediated
use. Additionally, familial processes may function
by elements of individuation, stress, marijuana use,
as risk or protective factors, either by driving ad-
deviance, and peer alcohol use. This ?gure is in-
olescents from a distasteful familial environment
tended as a heuristic model, not as a formal path
toward alcohol-using peer groups or by providing a
diagram to which partial regression coef?cients are
nurturing environment facilitative of healthy at-
assigned.
tachment to the family as individuation progresses.
The present study entails longitudinal prediction
Stress, use of other substances, deviant behavior,
of alcohol use among three groups of adolescents:
and peer alcohol usage are typically strong and
those who have never used alcohol (i.e., never
stable predictors of adolescents’ alcohol usage
users), those who have just begun to use alcohol or
(Donovan et al., 1991; Zucker, Boyd, & Howard,
are experimenting with it (i.e., initiators–moderate
1994).
experimenters), and those who consume heavily on
For example, family con?ict and parental use of
occasions of alcohol use (i.e., heavy users). Four
alcohol frequently predict adolescent alcohol use.
research questions are the focus of this investiga-
First, family con?ict is potentially indicative not
tion. First, do previously discovered robust predic-
only of a noxious family environment for the ad-
tors of adolescent alcohol usage in general function
olescent but also of parents who are often too
equally well as predictors of alcohol initiation or
distracted to communicate effectively with their
experimentation and of heavier alcohol usage?

---

104
GETZ AND BRAY
Figure 1.
Heavy alcohol use among adolescents as a function of family factors partially
mediated by individuation, prior marijuana use, stress, peer alcohol use, and deviance. Dashed
line paths dropped to nonsigni?cance after mediators entered the equation. T
time.
Second, are there unique predictors of heavier al-
measurement and whose alcohol consumption either re-
cohol usage? Third, do intrapersonal developmen-
mained the same or increased during the 3-year period.
tal and interpersonal psychosocial factors mediate
We imposed this restriction to eliminate the ambiguity
partially or completely the relationship between
associated with the alcohol status of claimed quitters or
more distal family processes and heavy alcohol use
consumption reducers. Mirroring the full longitudinal
among adolescents? Fourth, can it be demonstrated
sample (Baer & Bray, 1999), 49% of the 3,675-student
subset were male, 61% were members of intact families,
that robust risk and/or protective factors interact
and 39% were non-Hispanic White, 26% were African
synergistically such that negative consequences of
American, and 35% were Mexican American.
risk are buffered or mitigated (Rutter, 1985)? Al-
ternatively, can it be demonstrated that two co-
occurring protective factors produce a strong pre-
Instruments
ventive effect because one has potentiated the ac-
tion
of
the
other
(Brook,
Brook,
Gordon,
Demographic factors were the student’s gender, race/
Whiteman, & Cohen, 1990)?
ethnicity (non-Hispanic White, African American, and
Mexican American), age, and grade level; the mother’s
level of education; and the family structure (intact vs.
Method
nonintact).
Participants
Heavy alcohol use was measured in Year 3 by a single
?ve-point-scale item assessing the average quantity of
The
sample
for
this
study
was
selected
from
alcohol beverages consumed (i.e., beer, wine, and liquor
among 4,088 middle school-aged students who partici-
combined) during the previous month on an average
pated in a survey conducted annually in a cohort sequen-
occasion of use (1
nothing, I never drink, 5
six or
tial design in a large southwestern urban area over a
more drinks). This measure was then recoded as a tri-
period of 3 years (for a detailed description of the orig-
chotomy, such that three drinks or more indicated
inal study, see Baer & Bray, 1999). The 3,675 students
heavy use, less than one to two drinks re?ected ex-
who compose the study sample that we analyzed for this
perimental or moderate use, and nothing, I never drink
report were those who participated in all three waves of
re?ected abstention. The percentages of subjects in each

---

HEAVY ALCOHOL USE AMONG ADOLESCENTS
105
category of the trimetric code were 28%, 46%, and 26%,
alcohol and other substances, delinquency, and other prob-
respectively.1
lem behaviors. In contrast to separation, modalities of indi-
Family process and parental alcohol behavior were as-
viduation that have healthier consequences are represented
sessed with measures of family con?ict, mother’s monitor-
by the four other constructs we used. The entire Emotional
ing, communication with mother, and mother’s alcohol use.
Autonomy Scale has been validated and shown to be reli-
For all measures, higher mean scores indicated more of the
able (Steinberg & Silverberg, 1986). The Cronbach’s alpha
particular variable. The family con?ict measure was com-
value for the separation scale was .93 in our study sample.
posed of nine 5-point Likert items drawn from the Family
Intergenerational individuation is a component of the
Environment Scale (Moos & Moos, 1981), which indicates
Personal Authority in the Family System Questionnaire
openly expressed anger and aggression among family mem-
(Bray & Harvey, 1992; Bray et al., 1984) and is designed to
bers. The internal consistency (Cronbach’s alpha) for this
measure the extent to which adolescents and young adults
measure was .94 for the study sample, with 2-month test–
develop a sense of autonomy and self-determination while
retest reliability of .85 (Moos & Moos, 1981).
remaining emotionally connected to their parents. In this
Parental monitoring was assessed with a scale composed
context, individuation involves taking responsibility for
of seven 5-point Likert items that were drawn from the
oneself and not being controlled or impaired by one’s par-
adolescent version of the Assessment of Child Monitoring
Scale (Hetherington & Clingempeel, 1992). This scale mea-
sures the extent to which the mother is aware of the child’s
1Some researchers (Hays & Ellickson, 1996; Russell,
activities and friends, and it has been validated through
Welte, & Barnes, 1991) have suggested that global alcohol
factor analysis. It signi?cantly correlates with observations
use items (combining beer, wine, and liquor) and measures
of parent– child interactions on similar dimensions, and its
addressing either frequency or quantity of use, but not both,
test–retest reliabilities have ranged from .68 to .81. The
are less valid and/or reliable than measures that combine
Cronbach’s alpha value for our study sample was .81.
frequency and quantity and that assess separately beer,
Communication with mother was measured by ten 5-point
wine, and liquor consumption. Others (Windle, 1996) have
Likert items from the Barnes and Olson (1982) Parent–
highlighted the importance of qualitative distinctions in-
Adolescent Communication Scale. This scale assesses pos-
volving the mode of drinking (e.g., binging) or its conse-
itive aspects of parent–adolescent communication, includ-
quences (e.g., feeling drunk). For this reason, we con-
ing the free exchange of emotional and factual information,
structed a second, more complex indicator of heavy use. We
and has a 4-week test–retest reliability of .78. The Cron-
computed a general frequency indicator by summing the
bach’s alpha value for our study sample was .93.
scores of three beverage-speci?c (beer, wine, and liquor),
Mother’s perceived use of alcohol was measured with a
6-point Likert items; we adjusted this sum to weight for the
single 5-point item designed to assess the subject’s percep-
multiple ways a particular sum could be achieved at the low
tion of his or her mother’s drinking, ranging from not at all
end of the resulting 16-point indicator of frequency. Then,
(1) to every day (5).
we used scores on this frequency indicator to adjust sub-
Individuation/behavioral mediating factors were assessed
jects’ assignment to usage categories on the original quan-
with measures of ?ve constructs within the domain of
tity-based trichotomy as follows. Heavy users were de?ned
individuation and one measure of marijuana use. Separa-
as participants who consumed 3 or more drinks at one time
tion, Peer-Related Individuation (Levine, Green, & Millon,
within the past month and scored 8 or above on the fre-
1986), Intergenerational Individuation (Bray, Williamson,
quency indicator. Experimental/moderate users were de-
& Malone, 1984), Peer Trust (Armsden & Greenberg,
?ned as subjects who consumed 1 or more drinks at one
1987), and Parents’ Facilitation of Independence (Kenny,
time within the past month but scored between 3 and 7 on
1987) each capture different aspects of the individuation
the frequency indicator. Noninitiators/noncurrent users
process. Items on these scales factor distinctly from each
were de?ned as subjects who claimed they never drank and
other, and the respective factors they constitute are not
who scored 3 or less (never initiated or no consumption
highly correlated with each other. Higher mean scores on
within the past month) on the frequency indicator. The
each measure indicated more of the construct.
effect of this procedure was to more stringently de?ne
Separation was measured by nine items from the Stein-
heavy use, thus lowering the percentage of subjects so
berg and Silverberg (1986) Emotional Autonomy Scale.
designated from 26% to 14%. Percentages in each category
These items were the majority of those Steinberg and Sil-
of the quantity-based measure were as follows: never drink
verberg designated as indicative of nondependence on par-
(28%), less than 1 to 2 drinks per occasion (46%), and 3
ents and of parental deidealization. As noted by Ryan and
to 20 drinks per occasion (26%). Percentages in each cate-
Lynch (1989), the type of autonomy characterized by these
gory of the frequency by quantity indicator were as follows:
subscales might indicate some degree of alienation from
none (29%), experimental/moderate usage (58%), and
parents and a lack of con?dence in parental support and
heavy (14%). The two approaches to de?ning heavy alcohol
acceptance. The type of independence that develops in such
usage produced essentially the same results—which is not
a context has been characterized as emotional separation or
surprising because the two measures were very highly cor-
detachment. This modality of individuation, although it is
related (r
.89). Therefore, the presentation of data uses
functional in some respects, is correlated with the use of
only the more parsimonious, quantity-based measure.

---

106
GETZ AND BRAY
ents. This construct was measured by seven 5-point Likert
alpha value for this scale is .83; the value for our study
items. Two-week test–retest reliability for the Personal Au-
sample was .93.
thority in the Family System Questionnaire is .75 (Bray &
Harvey, 1992). The Cronbach’s alpha value for our study
Measurement Development
sample was .78.
Parents’ facilitation of independence was measured with
Because the measures described above were primarily
?ve of the eight items from Kenny’s (1987) Parental Rela-
developed for non-Hispanic White samples of adolescents,
tionship Questionnaire. This scale indicates the extent to
their reliability or external validity for our study could be
which the adolescent feels that his or her parents approve of
questioned. To make measures more comparable across
and facilitate his or her independent behavior and decision
attributes of gender and ethnicity, we subjected them to item
making. The Cronbach’s alpha value for our study sample
response theory (IRT) analysis (Thissen, 1991) and to con-
was .82.
?rmatory factor analysis. We used IRT methodology to
Peer trust was measured by 9 of the 10 items from the
develop comparable scales for each ethnic group by modi-
Inventory of Parent and Peer Attachment (Armsden &
fying or deleting some items that did not function equally
Greenberg, 1987). This scale assesses the extent to which
well across all racial/ethnic and gender categories. In addi-
the student feels that his or her friends are accepting, re-
tion, we established a response range of one to ?ve for items
spectful, and reliable con?dants. The Cronbach’s alpha
of any scale that originally had been coded with fewer than
value for our study sample was .93.
?ve ordinal attributes (Hulin, Drasgow, & Parsons, 1983) to
Peer-related individuation was measured by nine items
increase the potential information range of each scale. Com-
from the Healthy Separation scale, which was drawn from
parability among ethnic groups was improved as a result of
the Separation-Individuation Test of Adolescence (Levine
this analysis. Speci?cally, these procedures allowed the
et al., 1986). This scale assesses the extent to which ado-
conclusion that the information value provided by these
lescents are sure of themselves and are interpersonally at
scales did not vary substantively across ethnic groups or
ease with peers. The Cronbach’s alpha value for our study
between boys and girls.
sample was .77.
Other substance use (quantity of marijuana consumed)
Procedures
was measured with a single 5-point item ranging from 1
(none) to 4 (more than two joints for myself ) on an occasion
Sixth-, seventh-, and eighth-grade students from ?ve
of use within the previous year. Measured during the 2nd
school districts were surveyed in a classroom setting once
year of the study, this item is a marker for previous use of
each year for 3 years. One school district required active
other drugs and is a direct correlate of a subject’s alcohol
parental consent; the other four districts allowed passive
usage (r
.23). Psychosocial/behavioral mediating factors
consent—that is, parents gave consent by not precluding the
of stress, peer alcohol use, and deviant behavior are posited
participation of their child after receiving a letter announc-
as direct, positive predictors of adolescent alcohol use and
ing the survey date and purpose. Before administering the
as mediators of the impact of family-related behavioral
survey, one of our project staff members read a protocol to
factors and of individuation-related behavioral factors.
the students, explaining our purpose and indicating that
Higher mean scores on each scale indicated more of the
participation was voluntary. To preserve con?dentiality, our
construct.
staff conducted the survey and answered any questions that
Stress was measured with 13 items from Coddington’s
the students had, thereby avoiding contact between the
(1972) Life Events Scale for Children. This 5-point Likert
classroom teacher and the students during the period.
scale was structured to assess not only whether a stressful
event had occurred but also how much the occurrence
Data Analysis
bothered the respondent (Baer, Garmezy, McLaughlin,
Pokorny, & Wernick, 1987). The Cronbach’s alpha value
The four research questions were addressed in the context
was .76 for our study sample.
of hierarchic discriminate function analysis.2 Guided by the
Perceived peer alcohol use was assessed as the sum of
causal implications depicted in the psychosocial, develop-
two 5-point items adapted from Jessor and Jessor (1977).
One item asked about the alcohol consumption of “close
2Given that our dependent variable was dichotomous
friends”; the other asked about consumption among “kids
and our predictor variables were either dichotomous or
you know that are your age.” Possible scores on the result-
treatable as continuous, an alternative statistical approach
ing measure ranged from 2 to 10.
could be logistic regression. Because clear superiority of
Deviant behavior was measured during the 2nd year of
one technique over the other was not apparent in this in-
the study as the average of 16 ?ve-point scale items drawn
stance, we replicated the two-group, hierarchic entry ana-
from Jessor and Jessor’s (1977) longitudinal study of ado-
lytic procedure using logistic regression. Results were sub-
lescent problem behavior. The scale assesses the extent to
stantively similar to those using discriminant function anal-
which the student has engaged in antisocial or norm-violat-
ysis. Here we report only the discriminant function analysis
ing behavior within the previous year. The Cronbach’s
?ndings.

---

HEAVY ALCOHOL USE AMONG ADOLESCENTS
107
mental heuristic model (see Figure 1), we entered variables
tially mediated the linkage, we expected the parameter
sequentially in four blocks. Those within each block were
magnitudes of the family factor variables that were initially
forced into the equation rather than allowed to drop out as
signi?cant to drop substantially after the mediators entered
a result of not achieving an arbitrary level of statistical
the equation yet remain statistically signi?cant.
signi?cance. Each block contained six demographic vari-
Finally, we addressed our fourth hypothesis, that syner-
ables: sex, age, dummy codes for African American and
gistic effects might emerge with regard to the combining of
Mexican American ethnic status, intact family, and moth-
a risk and a protective variable (buffering), the combining of
er’s education level. These variables constituted the ?rst
two protective factors, or the combining of two risk factors.
block for two related reasons. First, the research questions
We combined the strongest risk factor discriminating heavy
involve the possible effects on heavy alcohol use of family
alcohol use (peer alcohol use) with the strongest inverse
factors, individuation/behavioral factors, and psychosocial–
discriminator (mother’s monitoring) to test the buffering
behavioral mediators when we control for or partial out the
hypothesis. A second test involved the combining of a
effects of gender, race/ethnicity, family structure, and pa-
second inverse discriminator (African American ethnic sta-
rental education (a proxy for family socioeconomic status
tus) with peer alcohol use. We tested protective factor
[SES]). Second, because racial/ethnic status is usually cor-
synergism by combining African American ethnic status
related with family structure and SES, any analysis using
with mother’s monitoring. We also tested risk factor syner-
racial/ethnic status as a predictor should include family
gism by combining separation with peer alcohol use and
structure and some measure of SES to minimize the possi-
with deviant behavior. For each test, we entered the inter-
bility that a signi?cant relationship involving racial/ethnic
action term into a trimmed model from which previously
status will be interpreted spuriously as causal.
nonsigni?cant variables had been eliminated. The interac-
The ?rst block contained the demographic variables only.
tion term was entered last, after its components and covari-
The second block contained family-related behavioral fac-
ates had been entered.
tors: family con?ict, mother’s alcohol use, mother’s moni-
toring, and communication with mother; these were mea-
Results
sured during the 1st year of the study. The third block added
the indicators of individuation: separation, intergenerational
The ?rst exploratory hypothesis was that variables
individuation, peer trust, parental facilitation of indepen-
previously identi?ed as either direct or inverse pre-
dence, and peer-related individuation; these were measured
dictors of adolescent alcohol initiation also discrim-
during the 2nd year of the study. Marijuana use was also
inate heavy use from experimental or moderate us-
included in the third block as a marker for previous illicit
age. Data addressing this hypothesis are presented in
drug use. The fourth block added the hypothesized psycho-
social and behavioral mediators, stress, peer alcohol use,
Table 1; here, 18 independent variables often associ-
and deviance, which were all measured during the 2nd year.
ated with adolescent alcohol use are blocked into one
The alcohol usage group status was based on the measure-
of the four categories noted in the text previously:
ments made during the 3rd year. In this way, we exploited
demographic factors, family process and parental al-
the longitudinal nature of the sample by matching our
cohol behavioral factors, individuation/behavioral
analytic procedure to the implied causal logic of the heu-
mediating factors, and psychosocial– behavioral me-
ristic model (see Figure 1), which distinguishes between the
diating factors. The literature suggests that discrimi-
impact of more distal family factors and the mediating
nators’ correlations with the function (displayed in
potential of more proximal, relational, or behavioral vari-
the structure matrix) constitute useful criteria for
ables (particularly those involving various dimensions of
assessing their importance (Klecka, 1980; Tabach-
the individuation process), and the psychosocial and peer-
related or behavioral alcohol involvement variables.
nick & Fidell, 1989). Discriminating variables with
To explore the ?rst two research questions involving the
signi?cant Wilks’s lambda ( p
.05) and Rao’s V
relative strengths of predictors, we conducted two analy-
values of .25 or greater were deemed worthy of
ses— one comparing the heavy use group with the experi-
attention. We regarded structure matrix parameters of
mental/moderate use group, and the other comparing the
.25 to .39 to be low, .40 to .59 to be moderate, and .60
experimental/moderate use group with the abstention group.
and above to be strong.
To explore the third question, regarding hypothesized me-
Correlations between the discriminating variables
diator effects, we compared parameter magnitudes and lev-
and the canonical discriminant functions (shown in
els of signi?cance of the variables entered sequentially in
Table 1) in cell entries for the model containing all of
the four blocks. If the variables intervening in the paths
the blocks of variables revealed that peer alcohol use
linking the family factor variables to heavy alcohol use (see
was, by far, the strongest predictor and a risk factor
Figure 1) were to wholly mediate the linkage, we expected
that those family factor variables that signi?cantly discrim-
for heavy use.
inated heavy from experimental/moderate alcohol use be-
Risk factors identi?ed as having moderate impor-
fore a mediator was entered into the equation would become
tance were deviant behavior and previous marijuana
nonsigni?cant after entry. If the intervening variables par-
use; those identi?ed as having low but notable im-

---

108
GETZ AND BRAY
Table 1
Structure Coef?cients/Canonical Function Coef?cients: Discrimination of Heavy Alcohol Users Compared
With Experimental/Moderate Users
Demog.
family factors
Demog.
family factors
individuation
factors
Demog.
Demog.
Demog.
individuation
psychosocial
trimmed
Discriminating variables
factors only
family factors
factors
mediators
2 (df) p
model
Demographic factors (T1)
Sex (1
male)
.39/.33***
.28/.23***
/.14*
/.13*
/.14***
African American
.59/ .69***
.43/ .54***
.35/ .44***
.30/ .38***
.30/ .38***
Mexican American
/.13*
/.14**
Intact family
/ .22*
Mother’s education
Age
.63/.67***
.45/.41***
.37/.30***
.32/.18***
140 (6) .001
.32/.19***
Family/behavioral factors
(T1)
Family con?ict
.47/.30***
.38/.20***
.32/.15**
.33/.10*
Mother’s monitoring
.61/ .50***
.50/ .30***
.42/ .18***
.43/ .14**
Communicate with
mother
.33/
/.15*
Mother’s alcohol use
.33/.28***
/.19***
/.11*
116 (4) .001
/.12*
Individuation/behavioral
mediators (T2)
Separation
.37/.23***
.32/
Intergenerational
individuation
Peer trust
Parents facilitate
independence
Individuation (peer)
Marijuana use
.69/.55***
.59/.28***
115 (6) .001
.59/.30***
Psychosocial/behavioral
mediators (T2)
Stress
Peer alcohol use
.73/.50***
.74/.51***
Deviant behavior
.58/.21***
127 (3) .001
.59/.20***
Note.
Structure coef?cients (to the left of the slash) are pooled within-groups correlations between discriminators and
standardized canonical discriminant functions. Empty cells re?ect parameters that had nonsigni?cant Rao’s Vs and
correlation values less than .25. Signs are adjusted to re?ect the direction of proportions or means relative to heavier alcohol
use. Alcohol use types after listwise deletion of missing data: heavy (n
661), experimental/moderate (n
1,179), nonuse
(n
672). Canonical function coef?cients are to the right of the slash. All p values refer to Wilks’s lambda. Demog.
demographic; T
time.
* p
.05.
** p
.01.
*** p
.005.
portance were separation and family con?ict. Age
Parameters to the left of the slash in the cells of
was also a risk factor of low importance. Potentially
Table 1 are not adjusted for multicolinearity among
important for understanding the role of previous mar-
the predictors. Parameters to the right of the slash are
ijuana use in alcohol initiation is the fact that the
standardized canonical discriminant function coef?-
adolescents who reported never having used mari-
cients that are adjusted for the presence of other
juana in the 2nd year had a 45% prevalence for
variables. The values in the fourth column indicate
alcohol use, whereas those who reported never hav-
that, except for separation, the risk and protective
ing used alcohol had only a 1% prevalence for mar-
factors noted above hold up as signi?cant discrimi-
ijuana use. The protective factors that were inversely
nators. The ?fth column in Table 1 presents the F and
related to heavy use as compared with experimental/
p values for each block of variables that succeeded
moderate use are mother’s monitoring (moderate
the ?rst block. Signi?cant standardized function co-
strength) and being African American (low strength).
ef?cients are displayed to the right of the slash in the

---

HEAVY ALCOHOL USE AMONG ADOLESCENTS
109
?fth column. Their signi?cance levels (based on
initiation of use equally well, or (c) neither heavy use
Wilks’s lambda) are indicated.
nor initiation of use. Table 3 displays structure matrix
The second exploratory hypothesis involved the
parameters in four columns representing these
possibility that some signi?cant discriminators of
possibilities.
heavy alcohol users, as compared with experimental
The second column in Table 3 indicates that family
or moderate users, would not necessarily be reliable
con?ict, marijuana use, being older, and being non-
discriminators between nonusers and experimental/
Hispanic White discriminated heavy users more reli-
moderate users (i.e., initiation) of alcohol (see Ta-
ably than initiators of alcohol use. Greater peer alco-
bles 1 and 2). Three other comparative possibilities
hol use and deviance and more effective maternal
are that a particular variable discriminates (a) exper-
monitoring discriminated heavy use and initiation
imental/moderate use from nonuse (i.e., initiation)
equally well, as noted in the ?rst column of Table 3.
more reliably than heavy use, (b) heavy use and
Better communication with mother, greater parental
Table 2
Structure Coef?cients/Canonical Function Coef?cients: Discrimination of Alcohol Experimental/Moderate
Users Compared With Never Users
Demog.
family
Demog.
factors
family factors
individuation
factors
Demog.
Demog.
Demog.
individuation
psychosocial
trimmed
Discriminating variables
factors only
family factors
factors
mediators
2 (df) p
model
Demographic factors (T1)
Sex (1
male)
African American
Mexican American
Intact family
.77/ .72***
Mother’s education
Age
.66/.62**
140 (6) .001
Family/behavioral factors
(T1)
Family con?ict
.38/
.29/
/.14*
Mother’s monitoring
.63/ .42***
.49/ .25***
.36/
Communicate with
mother
.57/.25
.44/
.32/
Mother’s alcohol use
.75/.67***
.58/.48***
.43/.31***
.43/.32***
Individuation/behavioral
mediators (T2)
Separation
.72/.53***
.53/.16**
.54/.22***
Intergenerational
individuation
.38/
Peer trust
Parents facilitate
independence
.46/
.34/
Individuation (peer)
Marijuana use
.40/.24***
.30/
115 (6) .001
Psychosocial/behavioral
mediators (T2)
Stress
.39/
Peer alcohol use
.81/.59***
.82/.61***
Deviant behavior
.64/.35***
127 (3) .001 .65/.39***
Note.
Structure coef?cients (to the left of the slash) are pooled within-groups correlations between discriminators and
standardized canonical discriminant functions. Empty cells re?ect parameters that had nonsigni?cant Rao’s Vs and
correlation values less than .25. Signs are adjusted to re?ect the direction of proportions or means relative to heavier alcohol
use. Alcohol use types after listwise deletion of missing data: heavy (n
661), experimental/moderate (n
1,179), nonuse
(n
672). Canonical function coef?cients are to the right of the slash. All p values refer to Wilks’s lambda. Demog.
demographic; T
time.
* p
.05.
** p
.01.
*** p
.005.

---

110
GETZ AND BRAY
Table 3
Comparison of Predictors’ Capacities to Discriminate Adolescent Heavy Users of Alcohol From
Experimental/Moderate Users and Experimental/Moderate Users From Nonusers: Structure Matrix
Correlations
Predicts relatively
Predicts heavy
Predicts initiation
Discriminating variables
equally
use better
better
Predicts neither
Demographic factors (T1)
Sex (1
male)
.20/.04
African American
.30/ .05
Mexican American
.06/.02
Intact family
.06/ .16
Mother’s education
.12/ .00
Age
.32/.18
Family/behavioral factors
(T1)
Family con?ict
.32/.22
Mother’s monitoring
.42/ .36
Communication with
mother
.23/ .32
Mother’s use of
alcohol
.23/.43
Individuation/behavioral
mediators (T2)
Separation
.32/.53
Intergenerational/individuation
.07/ .28
Peer trust
.07/ .02
Parents facilitate
independence
.10/ .34
Individuation (peer)
.10/ .07
Marijuana use
.59/.30
Psychosocial/behavioral
mediators (T2)
Stress
.24/.39
Peer alcohol use
.73/.81
Deviance
.58/.64
Note.
For the present purposes, to count as a substantive predictor, the correlation must be at least .25. Alcohol use types
after listwise deletion of missing data: heavy (n
661), experimental/moderate (n
1,179), nonuse (n
672). Heavy users
are to the left of the slash; experimental/moderate users are to the right. T
time.
facilitation of independence, and greater intergenera-
which variables were entered sequentially in accor-
tional individuation were more reliable inverse dis-
dance with their location in the pathways represented
criminators of initiation of alcohol use, whereas
in Figure 1.
mother’s frequent alcohol use, more stress, and
Table 1 presents standardized canonical coef?cient
greater separation predicted initiation moderately
values, whose signi?cance levels are shown to the
well and heavy use poorly. Peer trust and peer-related
right of the slash. Each entered block of measures
individuation predicted neither heavy use nor initia-
explained signi?cantly more variance at p
.001, as
tion of use; sex, intact family, and mother’s education
indicated by the data in the change in chi-square
level, at least in the context of the other variables in
column of Table 1. The trimmed model (column six
the model, also did not predict heavy use or initiation
of Table 1), in which individually nonsigni?cant or
of use.
low-parameter-magnitude measures were deleted
Our third hypothesis was that measures of individ-
from each block of discriminators, did not explain
uation, previous marijuana use, stress, peer alcohol
signi?cantly less variance than the full model,
2(7,
use, and deviance mediate relations between family-
N
1,840)
10.00, p
.10.
related behavioral factors and heavy alcohol use.
Assessing the mediating effect of each indicator of
Tests of this hypothesis are reported in Table 1.
individuation, we found that greater separation from
Because this aspect of the analysis was theory driven,
parents discriminated heavy alcohol use, even after
it was appropriate to use a hierarchic procedure in
the effects of family factor indicators were controlled

---

HEAVY ALCOHOL USE AMONG ADOLESCENTS
111
( p
.001). The other indicators within the domain of
Discussion
individuation were not signi?cant. Furthermore, the
parameter magnitudes for family con?ict ( p
.002),
In this study we found that certain family factors,
mother’s monitoring ( p
.001), and perceived ma-
the separation dimension of individuation, psycho-
ternal alcohol use ( p
.001) were consistent, weak-
social– behavioral factors, and ethnic status were im-
to-moderate discriminators of heavy use, which sug-
portant discriminators of adolescents who were
gests that they have direct effects, as well as indirect
heavy alcohol users when compared with those who
ones, mediated by separation. Marijuana use was a
were experimental or moderate users. In addition,
strong behavioral predictor of heavy alcohol use after
results reveal that other factors from those same
the family-related behavioral factors were controlled
domains were much stronger discriminators of exper-
( p
.001).
imental or moderate users than of nonusers of alco-
Entry of stress next and by itself into the model
hol. One family factor (mother’s parental monitoring)
(not shown in Table 1) did not affect the parameter
and two psychosocial behavioral factors discrimi-
magnitudes for separation, mother’s monitoring, or
nated membership in these groups equally well. Fi-
perceived maternal alcohol use. Stress was a signif-
nally, hypotheses regarding the mediation of the im-
icant predictor of heavy use (Wilk’s
.82, p
pact of family factors on alcohol usage by individu-
.001) before deviance and perceived peer use were
ation and psychosocial– behavioral factors were
entered into the model.
supported.
Entry of greater perceived peer alcohol use and
deviant behavior into the model resulted in a drop in
Direct Discriminators of Heavy
parameter magnitudes for family con?ict, mother’s
Alcohol Use
monitoring, perceived maternal alcohol use, and
stress as discriminators of heavy alcohol use. Each of
Four robust and moderately strong predictors of
the former variables except for stress remained a
heavy alcohol use among adolescents emerged: per-
signi?cant discriminator ( ps
.008, .003, and .023,
ceived peer alcohol use, race/ethnicity (i.e., being
respectively) in the presence of peer use and devi-
non-Hispanic White or Mexican American as op-
ance. This suggests that perceived greater use of
posed to African American), previous marijuana use,
alcohol by peers and deviant behavior partially me-
and age. Robust but weaker predictors were adoles-
diated the impact of the family factors. We suggest
cent deviant behavior, mother’s parental monitoring,
this possibility cautiously, as there is no test of sta-
mother’s alcohol use, and family con?ict. Of these
tistical signi?cance for the differences in parameter
weaker predictors, previous marijuana use and family
magnitude across blocks. Furthermore, perceived
con?ict were persistent discriminators only of heavy
peer use and deviance appeared to mediate com-
use as opposed to experimental/moderate use. In the
pletely the discriminating ability of separation and
context of the other predictor variables, the weaker
stress on heavy alcohol use, as they were no longer
predictors were not substantively important discrim-
signi?cant when peer use and deviant behavior were
inators of experimental/moderate use as opposed to
controlled. Marijuana use remained a discriminator
nonuse. Furthermore, age and being non-Hispanic
of heavy use, though of lesser magnitude ( p
.001).
White or Mexican American were both direct predic-
Although the demographic variables were not
tors of heavy use but not of experimental/moderate
study foci, we note that being African American
use. These ?ndings address the ?rst and second hy-
(compared with non-Hispanic White and Mexican
potheses; ?ndings here suggest that some well-estab-
American) remained a moderately strong protective
lished risk or protective factors are stronger or more
factor against heavy alcohol use ( p
.001), even
reliable markers for current or future heavy alcohol
after all other variables of the model were controlled.
use than they are for initiation of use.
Age, although it declined substantially in impact,
Why should previous marijuana use and family
remained a weak discriminator of heavy alcohol use.
con?ict operate in this way? With regard to mari-
Our fourth hypothesis was that risk and protective
juana use, this variable may well be a proxy for some
factors would interact with each other and produce
traits or processes that place adolescents at greater
synergistic outcomes. None of the interaction terms
risk for heavy alcohol use, such as use of multiple
used to test these hypotheses were found to be sta-
drugs, early age of initiation to use of alcohol or other
tistically signi?cant. Tests for interacting risk factors
drugs, and deviance or antisocial behavior (Windle,
and for interacting protective factors also were not
1996). The poor predictive capacity of marijuana use
found to be signi?cant.
for initiation of alcohol use suggests that marijuana

---