Content uploaded by Laurie Chassin
Author content
All content in this area was uploaded by Laurie Chassin
Content may be subject to copyright.
Binge Drinking Trajectories From Adolescence to Emerging Adulthood in
a High-Risk Sample: Predictors and Substance Abuse Outcomes
Laurie Chassin
Arizona State University
Steven C. Pitts
University of Maryland Baltimore Country
Justin Prost
Arizona State University
This study describes binge drinking trajectories from adolescence to emerging adulthood in 238 children
of alcoholics and 208 controls. Mixture modeling identified three trajectory groups: early–heavy (early
onset, high frequency), late–moderate (later onset, moderate frequency), and infrequent (early onset, low
frequency). Nonbingers were defined a priori. The early–heavy group was characterized by parental
alcoholism and antisociality, peer drinking, drug use, and (for boys) high levels of externalizing behavior,
but low depression. The infrequent group was elevated in parent alcoholism and (for girls) adolescent
depression, whereas the nonbinger and late–moderate groups showed the most favorable adolescent
psychosocial variables. All 3 drinking trajectory groups raised risk for later substance abuse or depen-
dence compared with the nonbingers, with the early–heavy group at highest risk.
Adolescence is the age period during which alcohol and drug
use are typically initiated. For example, national survey data
suggest that drinking onset and first intoxication usually occur
between Grades 7 and 10 (Johnston, O’Malley, & Bachman,
1999). Moreover, substance use shows systematic age-related
trends, with increases in consumption and abuse or dependence
diagnoses peaking in the age period that Arnett (2000) has called
“emerging adulthood” (i.e., ages 18–25; Chen & Kandel, 1995)
and then declining (Bachman, Wadsworth, O’Malley, Johnston, &
Schulenberg, 1997; Jessor, Donovan, & Costa, 1991). For exam-
ple, Bachman et al. (1997) found that increasing drinking after
high school was associated with leaving the parental home and
acquiring freedom from adult supervision, whereas declining
drinking between ages 22 and 32 was associated with entry into
marriage and parenthood. Thus, substance use increases are asso-
ciated with hallmarks of adolescent development such as individ-
uation and autonomy from adults, whereas substance use declines
are associated with hallmarks of emerging adulthood such as role
acquisition. Because of these age-related trends, substance abuse
has been referred to as a developmental disorder (Sher & Gotham,
1999; Tarter & Vanyukov, 1994).
However, simply because adolescent drinking is developmen-
tally normative does not mean that it is without clinical or public
health significance. Alcohol use can result in serious conse-
quences, such as injuries from impaired driving, and some adoles-
cents drink at problem levels. For the study of developmental
psychopathology, a critically important feature of adolescent alco-
hol use is its heterogeneity. That is, researchers must distinguish
drinking trajectories that are relatively benign from trajectories
that result in significant impairment.
Empirically modeling heterogeneity in drinking trajectories,
however, has been a somewhat elusive goal. Even growth model-
ing studies (e.g., Curran, Stice, & Chassin, 1997; Hussong, Curran,
& Chassin, 1998) have modeled individual differences in initial
levels of drinking and rates of escalation over time within the
context of a single overall group growth trajectory. These studies
do not address the potential heterogeneity of trajectory subgroups.
That is, they do not determine whether the overall group trajectory
contains within it different subgroups whose trajectories have
different shapes (as well as different antecedents or consequences;
Bates, 2000; K. G. Hill, White, Chung, Hawkins, & Catalano,
2000).
A few longitudinal studies have attempted to identify multiple
trajectories by cluster analyzing drinking patterns over time. Wills,
McNamara, Vaccaro, and Hirky (1996) found that a subgroup who
increased their alcohol and drug use between 7th and 9th grades
had high levels of stress, poor coping, low academic achievement,
low parental support, poor behavioral control, and parental and
peer substance use. Bennett, McCrady, Johnson, and Pandina
(1999) found that a subgroup of persistent problem drinkers from
age 18 to 31 were more likely to be disinhibited, male, and
characterized by problem behavior. Finally, Schulenberg, Wads-
worth, O’Malley, Bachman, and Johnston (1996) found that those
in trajectories of increased heavy drinking from age 18 to 24 were
more likely to be male, to be low in self-efficacy, and to drink in
order to get drunk.
Laurie Chassin and Justin Prost, Psychology Department, Arizona State
University; Steven C. Pitts, Psychology Department, University of Mary-
land Baltimore County.
This work was supported by Grant DA05227 from the National Institute
on Drug Abuse.
Correspondence concerning this article should be addressed to Laurie
Chassin, Psychology Department, Box 871104, Arizona State University,
Tempe, Arizona 85287-1104. E-mail: laurie.chassin@asu.edu
Journal of Consulting and Clinical Psychology Copyright 2002 by the American Psychological Association, Inc.
2002, Vol. 70, No. 1, 67–78 0022-006X/02/$5.00 DOI: 10.1037//0022-006X.70.1.67
67
NOTICE: This material may be protected
By copyright law (Title 17 U.S. Code)
Recent advances in mixture modeling provide a methodology to
empirically identify heterogeneity in trajectories over time by
clustering individuals according to their growth trajectories rather
than their raw scores (Muthen & Shedden, 1999; Nagin, 1999).
Advantages of this approach are that individuals do not have to be
measured at all occasions and that the measurement interval does
not have to be constant across individuals. Mixture modeling has
been applied to drinking in two studies. Muthen and Shedden
(1999) found that those who increased their binge drinking be-
tween age 18 and 24 were more likely to be diagnosed with alcohol
abuse or dependence compared with subgroups with either initially
high levels of binge drinking or low levels of binge drinking. K. G.
Hill et al. (2000) found four binge-drinking trajectories from
age 13 to 18: early high, increasers, late onsetters, and nonbingers.
Surprisingly, the groups did not differ on some earlier risk factors
such as internalizing symptoms, educational attainment, and fam-
ily bonding, although the nonbingers showed less problem behav-
ior. The trajectory groups diverged in their outcomes at age 21,
with the increasing and late onset groups having the highest rates
of alcohol abuse and dependence and the late onset group showing
the most drug abuse and dependence.
These studies are important in distinguishing subgroups whose
drinking is developmentally normative from those who develop
drinking problems. However, the few studies to examine multiple
drinking trajectories in adolescence also have limitations, which
the current study attempts to address. First, only one study (K. G.
Hill et al., 2000) spanned the years from alcohol initiation to the
ages of greatest risk for abuse and dependence diagnoses. Studies
of this age range are necessary to measure both early antecedents
and clinically relevant drinking outcomes. Otherwise it is not
possible to distinguish between normative and clinically signifi-
cant drinking trajectories in terms of their ages of onset and their
psychosocial antecedents. Second, none of these studies used a
high-risk sample, in which pathways to substance abuse and de-
pendence might be particularly evident. General population sam-
ples should show relatively low prevalence in problematic trajec-
tories, which can make it difficult to distinguish their antecedents
from those of more normative trajectory groups.
Given these limitations, the current study extends previous
research by examining multiple trajectories of binge drinking
using data from a longitudinal study of high-risk adolescents
(children of alcoholics, or COAs) and demographically matched
controls that spans age 12 to 23 (Chassin, Rogosch, & Barrera,
1991; Chassin, Curran, Hussong, & Colder, 1996; Chassin, Pitts,
DeLucia, & Todd, 1999). Methodological advantages include use
of a community sample, direct assessment of parental psychopa-
thology, low attrition, and multiple reporter data.
In addition to describing these trajectories, we asked whether
they could be prospectively predicted from well-established risk
factors for adolescent substance use. Following Sher’s (1991)
heuristic models of parent alcoholism effects on offspring alcohol
problems, we considered three categories of variables that differed
in how distal or proximal to adolescent drinking they are thought
to be. Parental psychopathology (alcoholism, antisocial personal-
ity, depression, anxiety disorder) was thought to be the most distal
risk factor. Given previous data, we expected parent alcoholism to
prospectively predict trajectories (Chassin et al., 1996; Sher,
1991). However, we also tested whether this effect was specific to
parent alcoholism (i.e., occurring above and beyond other diag-
noses) or whether parental impairment in general predicted binge-
drinking trajectories.
We also examined environmental risk factors, which have been
viewed as more proximal predictors of binge drinking. For exam-
ple, COAs experience higher levels of environmental stress and are
more likely to associate with peers who use alcohol (Chassin,
Barrera, & Montgomery, 1997). These adolescents are also more
likely to drink, perhaps to cope with stress (Wills, Vaccaro, &
McNamara, 1992) or because their peer networks provide drinking
opportunities and pro-drinking norms (Hawkins, Catalano, &
Miller, 1992). Accordingly, we examined stress and peer drinking
as environmental risk factors that might prospectively predict
trajectory group membership. Moreover, COAs are likely to be
exposed to family environment risk. COAs are more likely to be in
single-parent families, families with high levels of conflict, and
families with less consistent parental support and discipline (Chas-
sin et al., 1997). Such family environments have been shown to
raise risk for adolescent substance use and other problem behav-
iors (Hawkins et al., 1992; Sher, 1991). Accordingly, we tested
family structure and family environment risk (a composite of
family conflict, discipline, and parental support of the adolescent)
as prospective predictors of trajectory group membership.
We considered adolescents’ problem behaviors and symptom-
atology as the most proximal predictors of their drinking. That is,
the end result of exposure to multiple risk factors is hypothesized
to be an adolescent who is poorly regulated and undercontrolled
and who experiences high levels of negative affect (Sher, 1991).
These adolescents should exhibit higher levels of problem behav-
iors (i.e., externalizing symptoms and other drug use) and should
be more likely to follow problematic trajectories of binge drinking
as part of a broader pattern of “deviance proneness” (Jessor &
Jessor, 1977; Sher, 1991). Accordingly, we tested externalizing
symptoms and adolescent drug use as prospective predictors of
trajectory group membership.
The relation of negative affectivity to adolescent drinking is
more controversial (Swaim, Oetting, Edwards, & Beauvais, 1989).
Self-medication models suggest that adolescents drink to regulate
negative affect and thus that individuals with high levels of neg-
ative affect are at risk for binge drinking (Wills et al., 1992).
However, there has been mixed support for these models when
applied to adolescent drinking. Hussong et al. (1998) found that
internalizing symptoms did not predict growth over time in ado-
lescents’ drinking, and S. Hill, Shen, Lowers, and Locke (2000)
found that anxiety did not predict drinking onset in a high-risk
sample. Negative affectivity has been reported to be a stronger
predictor of problematic rather than normative adolescent drinking
(Colder & Chassin, 1999). Thus, negative affect might distinguish
between normative and problematic drinking trajectories even if it
does not predict drinking onset or escalation. Moreover, stronger
support for the role of negative affect might be produced by
disaggregating anxiety and depression, rather than using an “in-
ternalizing” measure. Some data suggest that adolescent depres-
sion is more strongly related to drinking than is anxiety (Hussong
& Chassin, 1993; Rohde, Lewinsohn, & Seeley, 1996). Accord-
ingly, we examined both adolescent anxiety and depression as
prospective predictors of trajectory group membership.
To help evaluate the clinical significance of these trajectory
groups, we compared their outcomes in terms of diagnoses in
emerging adulthood. Both K. G. Hill et al. (2000) and Muthen and
68
CHASSIN, PITTS, AND PROST
Shedden (1999) found that trajectories characterized by increasing
heavy drinking had higher prevalence of later alcohol diagnoses.
Moreover, given the correlation between alcohol and other drug
use, trajectories with more binge drinking should also show higher
rates of later drug diagnoses (K. G. Hill et al., 2000). Finally,
adolescent binge drinking is clinically important if it is associated
with later mental health problems. Given that adolescents who
drink heavily may experience negative consequences in terms of
their academic, psychological, and social functioning, they may be
more likely to develop anxiety or depressive disorders. Thus, we
tested whether trajectories with more binge drinking would show
more anxiety and depressive disorders in emerging adulthood.
In short, in the current study we aimed to describe multiple
trajectories of binge drinking from adolescence to emerging adult-
hood, to identify their prospective predictors, and (by examining
their diagnostic correlates in emerging adulthood) to ask which
trajectory groups had clinical implications for later outcomes.
Method
Participants
Participants were from an ongoing longitudinal study (Chassin et al.,
1991, 1996, 1999). At Time1 (mean age ⫽ 13.22 years, range ⫽ 10.5–
15.5), there were 454 adolescents, 246 of whom had at least one biological
and custodial alcoholic parent (COAs), and 208 demographically matched
non-COAs. There were three annual assessments and a follow-up 5–7 years
after Time 1.
Details of sample recruitment and representativeness are reported else-
where (Chassin, Barrera, Bech, & Kossak-Fuller, 1992). COA families
were recruited using court records of arrests for driving under the influence
of alcohol (n ⫽ 103), health maintenance organization questionnaires (n ⫽
22), and community telephone screening (n ⫽ 120). One family was
referred by a local hospital. COAs had to meet the following criteria:
Arizona residency, Hispanic or non-Hispanic Caucasian parents,
ages 10.5–15.5 years, English speaking, and no cognitive limitations that
would preclude interview. Direct interview data had to confirm that a
biological and custodial parent met Diagnostic and Statistical Manual of
Mental Disorders (3rd ed., DSM–III; American Psychiatric Association,
1980) criteria for lifetime alcohol abuse or dependence (using the Diag-
nostic Interview Schedule [DIS]; Robins, Helzer, Croughan, & Ratcliff,
1981) or Family-History Research Diagnostic Criteria (FH-RDC, Version
3; Endicott, Andreason, & Spitzer, 1975) using reports by the other parent
(if the alcoholic parent was not interviewed). At Time 1, 75.6% of biolog-
ical fathers and 86.6% of biological mothers were interviewed. If there
were multiple eligible children, the one closest to age 13 was selected.
Demographically matched controls were recruited using telephone in-
terviews. When a COA family was recruited, reverse directories were used
to locate families in the same neighborhood. Controls were screened to
match the COA in ethnicity, family structure, age (within 1 year), and
socioeconomic status (using the property value code from the reverse
directory). Interview data confirmed that neither biological nor custodial
parents met DSM–III criteria (or FH-RDC criteria) for lifetime diagnoses
of alcohol abuse or dependence. At Time 1, 71.2% of biological fathers
and 93.8% of biological mothers were interviewed.
Recruitment biases are discussed in detail elsewhere (Chassin et al.,
1992). The sample was unbiased with respect to alcoholism indicators
available in archival records (e.g., blood alcohol levels recorded at the time
of the arrest; see Chassin et al., 1992, for details). Moreover, the alcoholic
sample had rates of other psychopathology similar to those that have been
reported for a community-dwelling alcoholic sample (Helzer & Pryzbeck,
1988). However, those who refused participation were more likely to be
Hispanic, suggesting some caution in generalization.
A follow-up in emerging adulthood (hereafter referred to as Time 4) had
90% retention (n ⫽ 407; median age ⫽ 20.00 years, range ⫽ 18–23).
Retention was unbiased by gender and ethnicity, but slightly more COAs
(13.4%) than controls (6.7%) were lost,
2
(1, N ⫽ 454) ⫽ 5.45, p ⬍ .02.
The current analyses included those with data for at least three of the
four waves (n ⫽ 446; 98.2% of the sample). At Time 1, 52.9% were male,
average age was 13.23 years, 77.2% were non-Hispanic Caucasian, 81.2%
were living with two biological parents, and 65.5% had a parent with some
college education. The high percentage of two-biological-parent families
was produced by the selection criteria (i.e., requiring COAs to have both a
biological and a custodial alcoholic parent). COAs and controls did not
significantly differ, except that fewer COA parents had post-high school
education (57.6% of COAs vs. 74.5% of controls),
2
(1, N ⫽
446) ⫽ 23.44, p ⬍ .001.
Procedure
Data were collected with computer-assisted interviews either at families’
homes or on campus. To minimize contamination, family members were
interviewed individually, on the same occasion, by different interviewers
when possible. When a family moved out of state, an interviewer from a
nearby university administered a shortened version, and the diagnostic
interview was done by telephone. Interviewers were unaware of the fam-
ily’s group membership. Interviews required 1–2 hr and participants were
paid up to $65 over the waves.
To encourage honest responding, we reinforced confidentiality with a
Department of Health and Human Services Certificate of Confidentiality.
To maximize privacy, we gave participants the option of entering their
responses on the keyboard rather than making a verbal response.
Measures
All predictor variables were from Time 1 (except for parent anxiety
disorder, which was added only at Time 4; see Colder & Chassin, 1999, for
more details on reliability and validity of the predictors). Diagnostic
outcomes in emerging adulthood were from Time 4.
Parent alcoholism and other psychopathology. Parents’ lifetime
DSM–III diagnoses of alcoholism (abuse or dependence), affective disorder
(major depression or dysthymia) and antisocial personality disorder (ASP)
were obtained with a computerized version of the DIS (Version 3; Robins
et al., 1981). Parents’ lifetime DSM–III–R (Diagnostic and Statistical
Manual of Mental Disorders, 3rd ed., rev.; American Psychiatric Associ-
ation, 1987) anxiety disorder diagnoses (excluding simple phobia only)
were obtained with a computerized version of the DIS at Time 4. Diag-
noses were dichotomous variables either present (at least one biological
parent met lifetime criteria) or absent (neither biological parent met life-
time criteria).
1
Diagnoses in emerging adulthood. At Time 4, DSM–III–R diagnoses
of alcohol and drug abuse or dependence, ASP, affective disorder (major
1
Noninterviewed parents were considered not to meet criteria (except
for alcoholism, where FH-RDC was used for diagnosis from spousal
reports). This allowed us to include single-parent families, but it underes-
timates the prevalence of parental psychopathologies other than alcohol-
ism, which could produce negatively biased estimates of their effects. Note
that such underestimates could not occur when the interviewed parent met
diagnostic criteria, because in those cases parent psychopathology was
coded as present. Thus, these errors could occur only in cases where the
interviewed parent did not meet criteria and the noninterviewed parent
would have. Given our high interview rates for parents, this occurrence was
not frequent. On the basis of data from our two-interviewed-parent fami-
lies, estimates of potential misclassification errors were only 1% for
antisocial personality diagnoses and 3% each for depression and anxiety.
Thus, misclassification error should not substantially affect the findings.
69
SPECIAL SECTION: BINGE DRINKING TRAJECTORIES
depression or dysthymia), and anxiety disorder (excluding simple phobia
only) were made with the same computerized DIS measure. To ensure that
disorders were present after the adolescent measurement, we considered
only those who met both lifetime criteria and reported symptoms within the
past 5 years (for alcohol and drug diagnoses) or past year (for ASP,
anxiety, and depression) to manifest the disorder.
Environmental risk variables: Stress, peer drinking, and family environ-
ment. Stress was measured by adolescent and parent report of 15 uncon-
trollable negative events experienced by the child in the last 3 months
(from the General Life Events Schedule for Children; Sandler, Ramirez, &
Reynolds, 1986). Because there was sufficient agreement among reporters
(r ⫽ .57 between parents and .47 between parents and child), a multiple
reporter composite was used. Scores were standardized before aggregation
to avoid bias due to scaling.
2
Peer drinking was reported only by adolescents (because parents were
unlikely to have accurate knowledge). Two items (i.e., number of friends
who drank alcohol occasionally and regularly; response choices ranged
from 0 ⫽ none to 5 ⫽ all) were averaged for analysis (r ⫽ .87).
Family environment risk included family conflict, parental discipline,
and parental provision of social support. Adolescents and their parents
reported family conflict in the past 3 months using Bloom’s (1985) conflict
factor (e.g., “We fought a lot in our family”). Mother and father reports
were correlated (r ⫽ .47) and were averaged into an overall parent-report
measure.
Adolescents reported on parental consistent discipline using 10 items
from the Nonenforcement and Inconsistent Discipline subscales of the
Children’s Report of Parental Behavior Inventory (Schaefer, 1965; e.g.,
“My mother didn’t pay much attention to my misbehavior”). Because
adolescents’ ratings of their mothers and fathers were correlated (r ⫽ .72),
they were averaged. Parents’ self-reports used the same items. Because
mothers’ and fathers’ reports were not highly related (r ⫽ .15), mother
reports were treated separately (see Footnote 2).
Adolescents reported the support they received from their parents (seven
items from the Network of Relationships Inventory; Furman & Buhrm-
ester, 1985). Only adolescent reports were collected because support that is
not perceived by the adolescent was not viewed as helpful. Because their
ratings of mothers and fathers were highly correlated (r ⫽ .63), they were
averaged.
To reduce the number of predictors, we used confirmatory factor anal-
ysis to test whether adolescent and parent reports of conflict, adolescent
reports of discipline and social support, and mother report of discipline fit
a one-factor model. Mother report of discipline was not a good indicator
and was dropped from the composite and treated separately (
␣
⫽ .80). The
remaining measures fit a one-factor model,
2
(4) ⫽ 19.39, p ⬍ .05;
comparative fit index (CFI) ⫽ .95, and were used as a multiple reporter
measure of family environment risk (
␣
⫽ .91).
Adolescent symptomatology and problem behavior: Externalizing symp-
toms, drug use, depression, and anxiety. Adolescents and their parents
reported on 22 items from the Child Behavior Checklist (CBCL; Achen-
bach, 1978) that loaded on the externalizing factor for boys and girls ages
12–16. Because cross-reporter agreement was .65 between parents and .42
between parent and adolescent report, a multiple reporter composite was
used (
␣
⫽ .91).
Adolescents reported their frequency of drinking and drug use (eight
different drugs) with response options ranging from none to daily. The
frequency of consuming five drinks in a row in the past year was used to
create binge-drinking trajectories. The drug use predictor was the number
of drugs for which the adolescent reported any lifetime use at Time 1 (0 ⫽
none,1⫽ one,2⫽ two or more).
Adolescents and their parents reported on adolescents’ anxiety and
depression (in the past 3 months) with CBCL items. There were six
depression items (“lonely,”“cries a lot,”“feels unloved,”“worthless,”
“unhappy,” and “lacks energy”) and three anxiety items (“nervous,”“wor-
ried,” and “fearful/anxious”). There was low agreement among reporters
(e.g., for anxiety, rs ranged from .17 to .28), so adolescents’ reports (
␣
s ⫽
.80 for depression and .67 for anxiety; response options ranged from 1 ⫽
almost never to 5 ⫽ almost always) and mothers’ reports (
␣
s ⫽ .75 for
depression and .68 for anxiety; response options ranged from 1 ⫽ not true
to 3 ⫽ very often true) were tested separately.
Results
Empirically Identifying Binge Drinking Trajectories
Individual growth trajectories were clustered using mixture
modeling, a semiparametric, group-based approach (assuming that
a population contains different groups) that identifies between-
group differences in trajectory shapes (Muthen & Shedden, 1999;
Nagin, 1999). We used SAS PROC TRAJ (Jones, Nagin, &
Roeder, 1999) to model scores from Times 2–4 so that Time 1 risk
factors would be prospective predictors (after covarying Time 1
binge drinking). Trajectories were modeled as a function of age
rather than measurement wave. Given our design, this produced
fewer observations at the age extremes (e.g., a low of 23 obser-
vations at age 23) than at the middle of the age range (e.g., a high
of 124 observations at age 16).
A non-binge-drinking group (n ⫽ 176; 39.5% of the sample)
who reported no binge drinking at Times 2–4 was identified a
priori because their trajectory was known. The remaining 270
participants were used in the mixture modeling. Because binge
drinking was thought to increase during adolescence and then
stabilize or decline, we specified the highest order polynomial as
cubic. We first specified a single group and then tested a series of
models, increasing the number of groups and using the change in
the Bayes information criterion (BIC) to evaluate change in model
fit (Jones et al., 1999; Nagin, 1999). A three-group solution was
the most parsimonious fit to the data (change in BIC of 6.71 vs.
⫺11.20 for a four-group solution; a negative change in BIC
suggests a decrement in fit). Nonsignificant higher order polyno-
mial terms were trimmed. Support for the three-group solution was
seen in the average probability of group membership (.81; D.
Nagin, personal communication, 2001, March), and the fact that
only 9.2% of cases might be considered “difficult” to classify (i.e.,
had a probability of being assigned to a second group that was
above chance). Separate models were also estimated for males and
2
Because we wished to include single-parent families in analysis, our
approach was to first aggregate mother and father reports into an overall
“parent” variable whenever correlations allowed (i.e., correlations of .40 or
higher), using reports by the single parent if only one parent was present.
For all composite variables, scores were first standardized to avoid bias due
to scaling. However, when correlations between mother and father report
were less than .40, separate mother and father reports were maintained.
There were three predictors (parental discipline, adolescent anxiety, and
adolescent depression) for which agreement among reporters was insuffi-
cient to create multiple reporter constructs. In these cases, adolescent and
mother report were tested separately. Mothers’ reports were chosen over
fathers’ reports because the use of fathers’ reports requires that most of the
single-parent families be dropped from analysis, creating a nonrepresenta-
tive subsample and thus weakening the generalizability of the findings.
Consistent with this conclusion, other analyses of this data set (Chassin et
al., 1999) using imputation methods for missing parental diagnoses (Little
& Rubin, 1987) produced no differences in statistical inference and only
minimal changes in estimates of the coefficients.
70
CHASSIN, PITTS, AND PROST
females, but because the form and number of groups were quite
similar, an overall model was retained.
Binge drinking trajectory groups are shown in Figure 1. The
early–heavy group (n ⫽ 93; 20.9% of the sample, average prob-
ability of group membership ⫽ .85) were characterized by an early
onset of binge drinking, at about 13–14 years (i.e., the line for this
trajectory group left zero at this age, see Figure 1). They also
reached a high level of binge drinking, peaking at about 4.10
(binge drinking about weekly) at a relatively early age (19–20).
Although they showed some decline beyond this point, they still
remained fairly high relative to the other groups. The late–
moderate group (n ⫽ 134; 30.0% of the sample, average proba-
bility ⫽ .83) had a later onset of binge drinking (about 16–17) as
well as a lower maximum level (less than monthly). Finally, the
infrequent group (n ⫽ 43; 9.6% of the sample, average probabil-
ity ⫽ .76) was similar to the early–heavy group in having a
relatively early age of onset, but their binge drinking did not
escalate in frequency. Their maximum level was 0.90, where 1
represents binge drinking 1–2 times in the past year.
3
Adolescent Predictors of Binge Drinking Trajectory
Groups
To examine adolescent predictors of the trajectories, we com-
pared the four trajectory groups using discriminant function anal-
yses (DFAs). We tested three sets of risk factors: parent diagnoses,
environment risk, and problem behavior and symptomatology. For
each predictor, we report two effect sizes (f
2
; Cohen, 1988, where
values of .02, .15, and .35 represent small, medium, and large
effects). The unadjusted effect size is the relation between a
predictor and trajectory group membership unadjusted for any
other variables in the model, and the unique effect size is the
relation between a predictor and trajectory group membership
above and beyond the other predictors in the model. We computed
the unique effect size by comparing a full model with a reduced
model omitting the variable of interest.
4
For predictors whose
unique effect was significant at p ⬍ .05, we tested all pairwise
differences among the trajectory groups.
Time 1 covariates in the multivariate models were age, gender,
binge drinking, and family structure (living with both biological
parents vs. any other). Covariate by predictor interactions were
tested and probed using methods suggested by Aiken and West
(1990).
5
The first DFA tested the covariates only. The set of covariates
had a large effect in discriminating among trajectory groups, F(12,
1323) ⫽ 10.15, p ⬍ .01; f
2
⫽ .32, and all variables had significant
unadjusted relations to trajectory group membership. There were
significant unique effects of binge drinking, gender, and family
structure (see Table 1). Those with the most Time 1 binge drinking
were more likely to be in the early–heavy group than in any of the
others and in the infrequent group than in the nonbinger group (all
ps ⬍ .05). Males were more likely to be in the early–heavy and
late–moderate groups, whereas females were more likely to be in
the nonbinger and infrequent groups (all ps ⬍ .01). Finally, those
who were living with two biological parents were more likely to be
in the nonbinger or late–moderate group than in the early–heavy
group (ps ⬍ .05).
6
The next DFA tested parent diagnoses. This set of predictors and
their significant interactions with covariates discriminated among
the trajectory groups with a moderate effect size, F(15,
1308) ⫽ 4.31, p ⬍ .01; f
2
⫽ .16 (over and above the effects of the
covariates alone). Only parent alcoholism and ASP had significant
unadjusted and unique effects. Those with a parent diagnosed with
ASP were more likely to be in the early–heavy group than in any
of the others (ps ⬍ .05). Those with parent alcoholism were more
likely to be in any of the binge-drinking groups than in the
nonbinger group and more likely to be in the infrequent than in the
late–moderate group (ps ⬍ .05).
The effect of parental alcoholism was qualified by an interaction
with gender, F(3, 434) ⫽ 4.38, p ⬍ .01; f
2
⫽ .03, such that the
prediction of group membership from parental alcoholism was
stronger for females (f
2
⫽ .07) than for males (f
2
⫽ .04). Female
COAs were more likely to be in any of the binge drinking groups
than in the nonbinger group (40% COAs), and they were more
likely to be in the infrequent group (85%) than in the late–
moderate (65%) or early–heavy groups (65%, ps ⬍ .05). Male
COAs were more likely to be in the early–heavy group (75%
COAs) than either the nonbinger (39%) or late–moderate group
(45%, ps ⬍ .01).
The next DFA examined environmental risk (stress, peer drink-
ing, and family environment). These variables and their significant
3
To assess possible effects of missing data on trajectory group mem
-
bership, we compared the four groups with respect to the number of
members who were missing data at Time 4 (because there were virtually no
missing data between Times 1 and 3). As might be expected, because the
endpoint data are particularly important to defining their trajectory, the
late–moderate group had few cases that were missing Time 4 data (1%),
but the other three groups did not significantly differ (10% of early–heavy,
16% of infrequent, and 14% of nonbingers were missing at Time 4). We
also estimated the mixture model using only those with data from all waves
and replicated our trajectories. Thus, in the context of our very low attrition
rate (i.e., 90% of participants were retained through Time 4), attrition
should not have substantially influenced the findings.
4
To compute f
2
we used 1-Wilks’s lambda, which represents general
-
ized variance accounted for (Tatsuoka, 1988). The difference in Wilks’s
lambda values between the reduced model and the full model represents the
variance accounted for by the variable. Applying Cohen’s (1988) formula,
the effect size is computed as (Wilks’s lambda for the reduced model ⫺
Wilks’s lambda for the full model)/Wilks’s lambda for the full model.
Given the current sample size, a p level less than .05 corresponds to f
2
⫽
.018.
5
With the exception of gender, the interactions involving a specific
covariate were tested as a set. If the joint set of interactions involving the
covariate was significant, we then examined each specific interaction to
determine which predictor(s) interacted with the covariate. Interactions
involving gender were evaluated individually to provide a more sensitive
test, because our definition of binge drinking as consuming five drinks in
a row mades it more likely for males to meet the criterion. Although this
is a standard definition in the adolescent literature (Weingardt et al., 1998),
it is particularly important to test whether predictors of trajectory group
significantly differ for males and females because group membership
might have different meanings for males and females.
6
Although age had only a marginally significant unique effect on
trajectory group membership in the context of the other covariates, it did
have a significant unadjusted relation to trajectory group membership such
that younger participants were more prevalent in the nonbinger group. This
likely reflects the fact that their growth curves were projected on the basis
of data collected at younger ages. Accordingly, we retained Time 1 age as
a covariate in subsequent models.
71
SPECIAL SECTION: BINGE DRINKING TRAJECTORIES
interactions with covariates discriminated among the trajectory
groups with a moderate effect size, F(15, 1302) ⫽ 5.37, p ⬍ .01;
f
2
⫽ .20 (above and beyond the covariates alone). Although each
risk variable had a significant unadjusted relation to trajectory
group membership, only peer drinking had a significant unique
effect (see Table 1). Adolescents with greater peer drinking were
more likely to be in the early–heavy group than in either the
nonbinger or late–moderate group, and to be in the infrequent than
in the nonbinger group (ps ⬍ .05). However, the effect of peer
drinking was qualified by interactions with age, F(3, 432) ⫽ 4.69,
p ⬍ .01; f
2
⫽ .03, and gender F(3, 432) ⫽ 6.71, p ⬍ .01; f
2
⫽ .05.
The age interaction was such that younger individuals (1 SD below
the mean) mirrored the overall effects reported above, whereas
older individuals (1 SD above the mean) with more peer drinking
were more likely to be in the early–heavy group than in any of the
others (all ps ⬍ .05). The gender interaction was such that females
(f
2
⫽ .03) mirrored the overall effects reported above, whereas
males (f
2
⫽ .08) with more drinking peers were more likely to be
in the early–heavy group than in any of the others (ps ⬍ .01). We
also tested the environment risk model adding mothers’ report of
discipline as a predictor, but mother’s report of discipline did not
have a significant unique effect.
The final DFA examined adolescent symptomatology and prob-
lem behavior variables, which, along with their significant inter-
actions with covariates, discriminated among the trajectory groups
with a moderate effect size, F(24, 1299) ⫽ 4.41, p ⬍ .01; f
2
⫽ .27
(above and beyond the covariates alone). Each of the variables had
a significant unadjusted relation to trajectory group membership,
and all but anxiety showed significant unique effects (see Table 1).
Those with more externalizing symptoms were more likely to be
in the early–heavy group than in either the nonbinger or late–
moderate group (both ps ⬍ .01). However, these effects were
qualified by an interaction with gender, F(3, 431) ⫽ 3.13, p ⬍ .05;
f
2
⫽ .02, such that externalizing symptoms differentiated among
trajectory groups for males (p ⬍ .01, f
2
⫽ .03) but not females
(p ⫽ .54, f
2
⬍ .01). Males with more externalizing symptoms
were more likely to be in the early–heavy group than in either the
nonbinger or late–moderate group and more likely to be in the
infrequent than in the nonbinger group (all ps ⬍ .05).
Adolescents with more drug use were more likely to be in the
early–heavy or infrequent group than in either the nonbinger or the
late–moderate group (all ps ⬍ .05; see Table 1). However, these
effects were qualified by interactions with age, F(3, 431) ⫽ 3.62,
p ⬍ .05; f
2
⫽ .03, and gender, F(3, 431) ⫽ 3.17, p ⬍ .05; f
2
⫽ .02.
Considering the interaction with age, the pattern of significant
pairwise effects for younger participants was the same as that
shown in Table 1. However, for older participants, the only sig-
nificant pairwise difference was between the early–heavy and the
nonbinger groups. Considering the interaction with gender, males
(f
2
⫽ .03) with more drug use were more likely to be in the
early–heavy group than in either the nonbinger or the late–
moderate group (both ps ⬍ .01). Girls (f
2
⫽ .02) with high levels
of drug use were more likely to be in the infrequent than in either
the nonbinger or the late–moderate group and more likely to be in
the early–heavy than in the nonbinger group (ps ⬍ .05).
Finally, adolescent depression significantly differentiated the
trajectory groups such that increased depression was associated
with a greater likelihood of being in the infrequent group than in
the nonbinger group. Of interest, the early–heavy group showed
less depression than did the nonbinger, late–moderate, or infre-
quent group (all ps ⬍ .05; see Table 1). However, this effect was
qualified by a significant interaction with gender, F(3,
431) ⫽ 8.83, p ⬍ .01; f
2
⫽ .06. For males, (f
2
⫽ .06), decreased
depression was associated with a greater likelihood of being in the
early–heavy group than in the nonbinger, late–moderate, or infre-
quent group (all ps ⬍ .05), which did not differ from each other.
Thus, for boys, it was the early–heavy group that was distinct and
showed low depression. For females (f
2
⫽ .03), those with high
depression were more likely to be in the infrequent group than in
the nonbinger, late–moderate, or early–heavy group (all ps ⬍ .01),
which did not differ from each other. Thus, for girls, the infrequent
group was distinct and more depressed.
This model was also tested using mothers’ reports of the ado-
lescent’s anxiety and depression. Results replicated the adolescent
Figure 1. Growth curve trajectories of binge drinking from adolescence through emerging adulthood. Solid
lines represent estimated growth trajectories for the three groups from the mixture modeling. Dashed lines
represent observed means of binge drinking at each age for each group. Observed frequencies of binge drinking
(past year) ranged from 0 (none)to5(1–2 times a week). Nonbinger group, N ⫽ 176, 39.5% of the sample.
Early–heavy group, N ⫽ 93, 20.9% of the sample. Late–moderate group, N ⫽ 134, 30.0% of the sample.
Infrequent group, N ⫽ 43, 9.6% of the sample.
72
CHASSIN, PITTS, AND PROST
self-report model for the significant unique effect of depression,
with the early–heavy group showing less depression than the
late–moderate group and marginally less depression than the non-
binger group; however, there was no depression by gender inter-
action. As with adolescent self-reports, anxiety had no significant
unique effect.
Unique Effects of Predictors Across Models
To test whether the unique effects reported above would be
maintained above and beyond predictors from all of the models,
we tested a model in which all significant main effects and inter-
actions from the individual models were included (a total of 17
terms). The unique effects of parent alcoholism, externalizing
symptoms, and depression (and their interactions with gender)
were all maintained, but the unique effects of parental antisocial-
ity, Time 1 drug use, and peer drinking (and their associated
interactions) were no longer significant in this context.
Drinking Trajectories and Time 4 Emerging Adult
Outcomes
To evaluate the clinical significance of these trajectories, we
examined their relations to emerging adult diagnoses of alcohol
and drug abuse and dependence, affective and anxiety disorders,
ASP symptoms (there were too few diagnosed cases of ASP to
predict), and full-time college attendance. Logistic regressions
were used for dichotomous outcomes, and an ordinary least
squares regression was used to predict ASP symptoms. Covariates
(all from Time 1) were binge drinking, gender, age, family struc-
ture, and the relevant precursor of the outcome—that is, Time 1
drug use to predict Time 4 drug abuse and dependence, depressive
symptoms to predict affective disorder, externalizing symptoms to
predict antisociality, and anxiety symptoms to predict anxiety
disorder; no Time 1 analogue of college attendance was available.
The relevant parent diagnosis was also entered (i.e., parent alco-
holism to predict alcohol and drug abuse and dependence, parent
anxiety disorder to predict anxiety disorder, parent ASP to predict
antisociality, and parent depression to predict depression). Inter-
actions between trajectory group and each covariate were tested;
nonsignificant interactions were trimmed. If trajectory group had a
significant effect, then all pairwise drinking group differences
were tested.
In predicting alcohol abuse and dependence, males, those who
did not live with both biological parents, and COAs were at higher
risk for later alcoholism (all ps ⬍ .05). There was also a significant
effect of drinking trajectory group (see Table 2) such that the
early–heavy group had the greatest risk of diagnosis, and the
infrequent and late–moderate groups were at higher risk than the
Table 1
Effects of Adolescent Predictors on Binge Drinking Trajectory Group Membership
Variable and covariates
Observed
range
Unadjusted
effect size
Unique
effect size
F test of
unique effect
Nonbinger
(n ⫽ 176)
Late–moderate
(n ⫽ 134)
Infrequent
(n ⫽ 43)
Early–heavy
(n ⫽ 93)
T1 binge drinking 0–5
a
.17** .13 18.35** 0.05
a
0.13
a,b
0.42
b
0.88
c
T1 age in years 10.6–16.6 .04** .02 2.30 12.92 13.29 13.35 13.67
Gender (0 ⫽ female,1⫽ male)0–1 .08** .09 13.14** 0.40
a
0.62
b
0.37
a
0.72
b
Family structure 0–1 .04** .03 3.76* 0.12
a
0.17
a
0.26
a,b
0.31
b
Parental psychopathology risk
predictors
Alcohol DX
b
0–1 .09** .08 11.00** 0.39
a
0.51
b
0.77
c
0.72
b,c
ASP DX 0–1 .05** .02 3.06* 0.05
a
0.04
a
0.07
a
0.18
b
Depression DX 0–1 .01 ⬍.01 0.66 0.12 0.12 0.21 0.16
Anxiety DX 0–1 .01 .01 1.79 0.32 0.35 0.21 0.42
Environment risk predictors
Stress ⫺1.02–2.97
c
.06** .01 1.91 ⫺0.19 ⫺0.07 0.25 0.32
Family environment ⫺1.82–1.84
d
.09** ⬍.01 1.10 ⫺0.19 ⫺0.05 0.14 0.35
Peer drinking
b
0–5
e
.17** .06 8.45** 0.45
a
0.59
a,b
0.89
b,c
1.48
c
Problem behavior/symptomatology
risk predictors
Externalizing
b
⫺1.32–3.84
f
.15** .03 4.58** ⫺0.28
a
⫺0.07
a
0.22
a,b
0.52
b
T1 lifetime drug use
b
0–2 .14** .03 4.34** 0.06
a
0.12
a
0.26
b
0.54
b
Depression
b
1.00–4.50
g
.05** .03 4.80** 1.89
a
2.00
a,b
2.50
b
1.98
c
Anxiety 1.00–5.00
g
.02* ⬍.01 0.30 2.23 2.31 2.70 2.41
Note. Unadjusted effect size refers to the relation between the predictor and trajectory group membership unadjusted for any other variables in the model.
Group means that do not share a subscript differ significantly at p ⬍ .05. Tests of pairwise group differences were performed only if the predictor had a
significant unique effect (p ⬍ .05) on trajectory group in the model including all hypothesized predictors and any predictor by covariate interactions.
Degrees of freedom for the unique F tests were (3, 439) for the covariates only, (3, 434) for the parent psychopathology model, (3, 432) for the environment
risk model, and (3, 431) for the problem behavior model. Raw means are presented, but results, including pairwise comparisons, are based on the model
including all hypothesized predictors and any Predictor ⫻ Covariate interactions. T1 ⫽ Time 1; DX ⫽ diagnosis; ASP ⫽ antisocial personality disorder.
a
Scores range from 0 (no past year binge drinking)to5(1–2 times per week).
b
This predictor interacted with one or more covariates in predicting
drinking trajectory group. Pairwise effects listed in the table are the effects of the predictor at the average level of the covariate(s).
c
Composite score that
was standardized within reporter and then averaged.
d
Composite score from multiple measures and reporters that were standardized within reporter and
then averaged.
e
Scores range from 0 (none)to5(all).
f
Composite score that was standardized within reporter and then averaged.
g
Scores range
from 1 (almost never)to5(almost always).
* p ⬍ .05. ** p ⬍ .01.
73
SPECIAL SECTION: BINGE DRINKING TRAJECTORIES
nonbingers (all ps ⬍ .05; see Table 2). However, this effect was
qualified by an interaction with gender,
2
(3, N ⫽ 404) ⫽ 12.14,
p ⬍ .01, such that the effect of drinking trajectory group was larger
for males,
2
(3, N ⫽ 404) ⫽ 65.86, p ⬍ .01, than for females,
2
(3,
N ⫽ 404) ⫽ 19.02, p ⬍ .01. For males, being in any of the
binge-drinking groups increased risk for alcohol diagnoses (44%
diagnosed for late–moderate, 69% for infrequent, and 84% for
early–heavy) compared with nonbingers (5%), and the early–
heavy group was at higher risk than the late–moderate group (all
ps ⬍ .01). For females, being in the early–heavy group increased
risk for alcohol diagnoses (73% diagnosed) relative to all other
groups (all ps ⬍ .01), which did not significantly differ from each
other (nonbingers ⫽ 13%, late–moderate and infrequent groups
both ⫽ 30%).
In predicting drug abuse and dependence, those who were
younger, who did not live with both biological parents, who had
more Time 1 drug use, and who were COAs were more likely to
be diagnosed (all ps ⬍ .05). There was also a significant effect of
trajectory group. Those who were in any of the binge-drinking
trajectories (compared with the nonbingers) were more likely to be
diagnosed with drug abuse and dependence (all ps ⬍ .01; see Table
2).
In predicting ASP symptoms, those who were younger, males,
and those with more Time 1 externalizing had more symptoms (all
ps ⬍ .05). There was also a significant effect of trajectory group.
Those in the early–heavy group had more symptoms than any of
the others (ps ⬍ .01), and those in the late–moderate group had
more symptoms than the nonbingers (p ⬍ .05; see Table 2).
In predicting depression diagnoses, females and those with more
Time 1 depression were more likely to be diagnosed (both ps ⬍
.01). There were no significant effects of trajectory group. Testing
the model using mothers’ report of adolescent depression repli-
cated these findings.
In predicting anxiety disorders, those who did not live with both
biological parents at Time 1 and those with more Time 1 anxiety
were more likely to be diagnosed (both ps ⬍ .05). Those with
parental anxiety disorder were marginally more likely to be diag-
nosed (p ⬍ .09). There were no significant effects of drinking
group. Testing the model using mothers’ reports of the adoles-
cent’s anxiety replicated these effects and produced a significant
effect of gender (p ⬍ .05) such that females were more likely to
be diagnosed than were males.
In predicting college attendance, there were no significant ef-
fects of covariates, but there was an effect of trajectory group.
Those in the nonbinger group were more likely to be in college full
time than were those in any of the other groups (all ps ⬍ .01; see
Table 2).
7
Discussion
The current study identified multiple trajectories of binge drink-
ing from adolescence to emerging adulthood and distinguished
among them in terms of both their antecedents and their clinical
significance for diagnoses in emerging adulthood. Specifically, we
identified a group of nonbingers who reported never having drunk
five drinks in a row, a group whose binge drinking began early in
adolescence and increased to high levels (early–heavy), a group
whose binge drinking began early but stayed at occasional levels
of frequency (infrequent), and a group whose binge drinking began
later in adolescence and increased to moderate levels (late–
moderate). These trajectories are similar to those identified by
K. G. Hill et al. (2000) with a somewhat younger sample, who
found a nonbinging group, an early onset but not increasing group
(similar to our infrequent group), an increasing group (similar to
our early–heavy group), and a later onset group (similar to our
late–moderate group. The similarity in these empirically identified
7
For this model, we selected only participants who were 19 and older to
ensure that they were old enough to have completed high school. Those
who reported being full-time students and also reported attaining at least
some college education could thus be confidently viewed as college stu-
dents rather than high school students.
Table 2
Effect of Trajectory Group Membership on Emerging Adult Outcomes
Variable Range
a
Unadjusted effect
size of group
Unique effect
size of group
2
test of
unique effect
Nonbinger
(n ⫽ 152)
Late–moderate
(n ⫽ 132)
Infrequent
(n ⫽ 36)
Early–heavy
(n ⫽ 84)
Alcohol abuse
b
0–1 .46** .30 71.95** 0.10
a
0.39
b
0.44
b
0.81
c
Drug abuse 0–1 .09** .05 18.12** 0.03
a
0.17
b
0.25
b
0.27
b
ASP symptoms
c
0–10 .15** .07 F ⫽ 9.34** 0.32
a
0.87
b
0.69
a,b
1.62
c
Affective disorder 0–1 .01 .02 6.53 0.05 0.09 0.11 0.13
Anxiety disorder 0–1 .01 .01 4.10 0.11 0.17 0.25 0.13
College attendance
d
0–1 .06** .06 17.33** 0.44
a
0.22
b
0.16
b
0.21
b
Note. Unadjusted effect size of group refers to the relation between trajectory group membership and the outcome unadjusted for any other predictors in
the model. For each test, N ⫽ 404 and df⫽3, with the exception of antisocial personality disorder (ASP) symptoms (F tests in which dfs were [3, 394] and
college attendance (N ⫽ 323, df ⫽ 3). Group proportions that do not share a subscript differ significantly at p ⬍ .05. Pairwise comparisons were conducted
only if trajectory group had a significant unique effect in the model including all covariates and any Trajectory Group Covariate interaction. Raw
proportions are presented, but results of pairwise comparisons are based on the model including all covariates and any Trajectory Group ⫻ Covariate
interaction. ASP ⫽ antisocial personality disorder
a
Prevalence of diagnosed cases, with the exception of ASP symptoms and full-time college attendance (0 ⫽ not attend,1⫽ attend).
b
Drinking group
interacted with gender in the prediction of alcohol abuse. Pairwise effects listed are the effects of group at the average level gender (weighted effect
codes).
c
Model predicting ASP symptoms was estimated with ordinary least squares multiple regression. The test of the unique effect of group was a
change in R
2
test in which dfs were (3, 395).
d
Sample sizes for this analysis were 109, 111, 31, and 72 for the Nonbinger, Late–moderate, Infrequent,
and Early–heavy groups, respectively.
** p ⬍ .01.
74
CHASSIN, PITTS, AND PROST
trajectory groups across two disparate samples increases confi-
dence in their validity. Moreover, not only did these groups differ
in age of onset and in the shape of their trajectories, they also
differed in their antecedents (with the early–heavy and infrequent
groups generally showing the highest levels of risk factors) and in
their clinical significance (with all the binge-drinking trajectories
showing increased risk for alcohol and drug abuse and dependence
compared with the nonbinger group, and the early–heavy group
showing the greatest risk).
From a clinical perspective, the most worrisome group was the
early–heavy group, whose binge drinking started in early adoles-
cence and occurred at high levels of frequency. Because of their
increased frequency of binge drinking, they are likely to be at
greatest risk for short-term negative consequences, and they were
also at the highest risk to develop young adult alcohol abuse and
dependence. This trajectory should not be viewed as developmen-
tally normative drinking because these adolescents are likely to
develop levels of social consequences and dependence symptoms
that warrant clinical diagnoses.
In adolescence, the early–heavy group showed elevations in
multiple risk factors, including parent alcoholism and antisociality,
adolescent drug use and (for males) higher levels of externalizing
symptoms. These data replicate findings from studies of marijuana
use (Kandel & Chen, 2000) and cigarette smoking (Chassin, Pres-
son, Pitts, & Sherman, 2000) in identifying a subgroup with both
early onset and high levels of use as most problematic, both in
terms of adolescent risk factors and later heavy use. In addition,
this subgroup resembles previous descriptions of a subtype of
alcoholism (identified among adult alcoholics) that is more com-
mon in males and characterized by early onset and high levels of
antisociality (Babor et al., 1992; Cloninger, 1987). Although stud-
ied in adulthood, this subgroup has also been reported to show high
levels of childhood risk factors (Babor et al., 1992). Interestingly,
the early–heavy group cannot be simply dismissed as the most
“risky” in general, because (particularly for boys) they showed
significantly less depression than did any of the other groups.
These findings suggest that negative affect regulation motives may
not underlie their early and steep rise in binge drinking and that
drinking for this group is more closely tied to “deviance prone-
ness” pathways characterized by a broader pattern of externalizing
behavior (Sher, 1991).
The infrequent group also showed early onset of binge drinking,
although they did so at lower levels of frequency than the early–
heavy group. Despite their low frequency of binge drinking, the
infrequent group should not be considered a group without risk.
Although they were less likely than the early–heavy group to
develop alcoholism or antisociality, they were as likely as the
early–heavy group (and more likely than the nonbinger group) to
develop drug abuse and dependence.
8
Of note, in terms of adoles-
cent risk factors, some of the elevated risk associated with mem-
bership in the infrequent group was specific to females. For girls
but not boys, there was more parental alcoholism in the infrequent
than in the early–heavy group, and for girls but not boys, it was the
infrequent group that was distinguished from all of the others in
terms of elevated depression. In addition, there were more females
than males within this trajectory group. Taken together, these
findings suggest that the infrequent group represents a binge-
drinking trajectory that is more characteristic of females than
males and whose drinking is more tied to negative affect regula-
tion, specifically to elevated levels of depression. This group is
reminiscent of Cloninger’s (1987) Type I alcoholism, in having
large numbers of females who are high in negative affect. How-
ever, Cloninger (1987) described this group as showing alcoholism
onset after age 25, in contrast to our infrequent group, which began
binge drinking at an early age and for whom 44% were diagnosed
with alcohol abuse or dependence by age 23.
Also consistent with previous literature on other forms of sub-
stance use was our identification of a group of late onset binge
drinkers who showed moderate frequency (Chassin et al., 2000). In
adolescence, this late–moderate group differed from the early–
heavy group in many ways—showing less parent alcoholism and
antisociality, less peer drinking (for older participants), less drug
use (for younger participants), and (for boys) less externalizing
behavior. The onset of their binge drinking coincided with the end
of their high school years. This later age of onset suggests that
binge drinking for this group might represent more developmen-
tally normative “partying” that occurs when adolescents are less
subject to parental supervision (i.e., after the ages at which they
can drive or transition out of the parental home). Although such
increases have often been associated with pro-drinking social
norms of college environments, in our data, the late–moderate
group was not particularly elevated in full-time college attendance.
Again, however, this group is not entirely without risk. Although
they were not at elevated risk for depression or anxiety disorder,
they were more likely to develop substance use disorders than
were their peers who did not engage in any binge drinking. Thus,
whether or not binge drinking is correlated with earlier adolescent
risk factors, and even if it occurs more in a context that is more age
typical, there can still be risk for the development of problematic
levels of abuse and dependence, perhaps simply as a result of
repeated consumption.
In contrast to these groups, the nonbingers, who did not engage
in binge drinking, generally had the most favorable pattern of
adolescent psychosocial variables and the least risk for later alco-
hol and drug diagnoses. In addition, those without adolescent
binge drinking were the most likely to be full-time college students
in emerging adulthood. This is consistent with a role selection
effect in that those who were best adjusted and had fewest ado-
lescent risk factors are most likely to select (and to be selected)
into higher education (Bachman et al., 1997). Thus, although some
adolescent drinking may be normative, our data suggest that binge
drinking is less benign and that the most favorable developmental
trajectory is abstaining from binge drinking.
These data may appear to conflict with Shedler and Block
(1990), who suggested that adolescent abstinence from marijuana
use was associated with poorer psychosocial adjustment. However,
Shedler and Block (1990) found that frequent marijuana use also
8
One possible interpretation is that although the infrequent group en
-
gages in binge drinking at a low level of frequency, they engage in other
forms of adolescent substance use at very high levels, and that is why they
are elevated on adolescent risk factors and drug diagnoses in emerging
adulthood. We tested this possibility by examining the frequency of mar-
ijuana use (the most commonly used substance in our data set) as well as
simple frequencies of drinking (rather than consuming five drinks in a row)
at each wave of measurement. However, the infrequent group did not show
any particular elevation or escalation in marijuana use, beer or wine use, or
hard liquor use that would support such an interpretation.
75
SPECIAL SECTION: BINGE DRINKING TRAJECTORIES
was associated with poor adjustment, and our binge drinking
variable may be more comparable to their frequent marijuana use
than to experimental use. Moreover, other studies (Milich et al.,
2000; Wills et al., 1996) have found the most positive adjustment
to be characteristic of abstainers, suggesting that Shedler and
Block’s findings may not generalize to more recently measured
cohorts.
The current findings have implications for the three sets of risk
factors (parent psychopathology, risky environments, and adoles-
cent symptomatology) and for deviance proneness and negative
affect regulation theories of adolescent drinking. As hypothesized,
parent psychopathology did prospectively predict binge drinking
trajectories. Moreover, these effects were unique to parent alco-
holism and antisociality rather than attributable to parent impair-
ment in general. Consistent with previous theory and data (Chassin
et al., 1996; Zucker, Boyd, & Howard, 1994), parent alcoholism
was associated with elevated risk for membership in all of the
binge-drinking groups (particularly the early–heavy group for
boys and the infrequent group for girls), and parent antisociality
was associated with membership in the group that had the earliest
onset and highest frequency of binge drinking (i.e., the early–
heavy group). Moreover, although some of the parent alcoholism
effect might be mediated through risky environments and adoles-
cent symptomatology, these variables did not eliminate the parent
alcoholism effect on trajectory group membership. This suggests
that other mediators are required to explain parent alcoholism
effects. For example, perhaps COAs derive different pharmaco-
logical effects from drinking, and these effects then influence their
binge-drinking trajectories (Schuckit, 1994; Sher, 1991).
In terms of risky environments, our findings replicate previous
data in suggesting that stress (Wills et al., 1996), single-parent
families, and family environments that are high in conflict but low
in discipline and nurturance (Hawkins et al., 1992) prospectively
predict adolescent alcohol involvement. However, except for fam-
ily structure, these effects were no longer significant in multivar-
iate models after peer drinking was entered. Although a formal test
of mediation is beyond the scope of this article, these findings are
consistent with a model in which the effects of stress and family
conflict, support, and discipline on adolescent drinking are medi-
ated through peer influences. Other studies have suggested that
stress and poor socialization result in affiliation with a peer social
network that models and reinforces drinking. This peer network is
a proximal mediator of adolescent drinking (e.g., Dishion, Patter-
son, & Reid, 1988).
Finally, the current findings have important implications for
deviance proneness and negative affect regulation models of ado-
lescent drinking. The findings add support to the often replicated
result that conduct problems (here including externalizing symp-
toms and early drug use) describe a subgroup of adolescents who
initiate drinking at an early age and escalate quickly to problematic
levels with an elevated risk for alcohol abuse and dependence
(Sher, 1991; Zucker et al., 1994). More important, however, our
ability to distinguish among heterogeneous trajectories of binge
drinking revealed support for the more controversial and less
replicated negative affect regulation pathway. Because this support
was confined to females and showed elevations for the infrequent
group compared with the other trajectories, it would not have been
revealed in conventional analyses predicting the frequency of
binge drinking, even in growth models predicting individual dif-
ferences in increases in binge drinking over time. The fact that
depression rather than anxiety had a unique effect on trajectory
group membership is consistent with other data finding stronger
effects of adolescent depression than anxiety in predicting drinking
(e.g., Costello, Erkanli, Federman, & Angold, 1999; S. Hill et al.,
2000) but may also be due to more reliable measurement of
depression than anxiety in the current data set.
The finding of lowered depression for males in the early–heavy
group is counterintuitive and requires replication. Perhaps these
boys are drinking in a peer social context that reduces their
depressive affect. Of interest, Labouvie, Pandina, White, and John-
son (1990) hypothesized that one pathway underlying adolescent
substance use is characterized by low levels of constraint but high
levels of positive affect, and this combination is similar to the high
level of externalizing symptoms and low levels of depression
among males in our early–heavy group. It is important to note,
however, that their lowered levels of depression were not main-
tained into emerging adulthood, at which point the early–heavy
group did not show lowered prevalence of depression diagnoses.
This may reflect an effect of heavy drinking and its associated
consequences, which could serve to increase subsequent depres-
sion (Swendsen & Merikangas, 2000).
Although the current study extended previous work by exam-
ining a high-risk sample over a large age range, by directly
assessing diagnoses of parents and offspring, and by using multiple
reporter data, it is also important to consider some of its limita-
tions. First, although the sample size is relatively large for COA
research, the sizes of the trajectory groups did not provide optimal
statistical power to detect interaction effects and pairwise group
differences. Moreover, our sample sizes are smaller at the age
extremes of the trajectories than in the middle range of ages, where
more data points are available. Second, although consuming five
drinks in a row is a common definition of adolescent binge
drinking, different findings can be produced by different defini-
tions, and the predictive utility of this definition may vary for
males and females (Weingardt et al., 1998). Third, our measures of
adolescent depression and anxiety were brief, and more compre-
hensive measures (particularly diagnosis rather than symptomatol-
ogy) might produce different findings. For example, it has been
suggested that social phobia is a predictor of substance abuse and
dependence (Merikangas & Avenevoli, 2000), and such an effect
would not be detected in our study. Finally, because we followed
participants only to age 23, we are better able to model the
age-related increases in drinking that occur from adolescence to
emerging adulthood than the declines that are likely to occur after
that age. Further follow-ups are needed to better illuminate heter-
ogeneity in the adult years.
References
Achenbach, T. M. (1978). The child behavior profile: I. Boys 6–11.
Journal of Consulting and Clinical Psychology, 47, 223–233.
Aiken, L., & West, S. (1990). Multiple regression: Testing and interpreting
interactions. Newbury Park, CA: Sage.
American Psychiatric Association. (1980). Diagnostic and statistical man-
ual of mental disorders (3rd ed.). Washington, DC: Author.
American Psychiatric Association. (1987). Diagnostic and statistical man-
ual of mental disorders (3rd ed., rev.). Washington, DC: Author.
Arnett, J. J. (2000). Emerging adulthood: A theory of development from
76
CHASSIN, PITTS, AND PROST
the late teens through the twenties. American Psychologist, 55,
469–480.
Babor, T., Hoffman, M., DelBoca, F., Hesselbrock, V., Meyer, R., Dolin-
sky, Z., et al. (1992). Types of alcoholics: I. Evidence for an empirically
derived typology based on indicators of vulnerability and severity.
Archives of General Psychiatry, 49, 599–608.
Bachman, J. G., Wadsworth, K., O’Malley, P., Johnston, L., & Schulen-
berg, J. (1997). Smoking, drinking and drug use in young adulthood: The
impacts of new freedoms and new responsibilities. Mahwah, NJ:
Erlbaum.
Bates, M. E. (2000). Integrating person-centered and variable-centered
approaches in the study of developmental courses and transitions in
alcohol use: Introduction to the special section. Alcoholism: Clinical and
Experimental Research, 24, 878–881.
Bennett, M. E., McCrady, B., Johnson, V., & Pandina, R. J. (1999).
Problem drinking from young adulthood to adulthood: Patterns, predic-
tors, and outcomes. Journal of Studies on Alcohol, 60, 605–614.
Bloom, B. (1985). A factor analysis of self-report measures of family
functioning. Family Process, 24, 225–239.
Chassin, L., Barrera, M., Bech, K., & Kossak-Fuller, J. (1992). Recruiting
a community sample of adolescent children of alcoholics: A comparison
of three subject sources. Journal of Studies on Alcohol, 53, 316–320.
Chassin, L., Barrera, M., & Montgomery, H. (1997). Parent alcoholism as
a risk factor. In S. Wolchik & I. Sandler (Eds.), Handbook of children’s
coping (pp. 101–130). New York: Plenum.
Chassin, L., Curran, P., Hussong, A., & Colder, C. (1996). The relation of
parent alcoholism to adolescent substance use: A longitudinal follow-up
study. Journal of Abnormal Psychology, 105, 70–80.
Chassin, L., Pitts, S., DeLucia, C., & Todd, M. (1999). A longitudinal
study of children of alcoholics: Predicting young adult substance use
disorders, anxiety, and depression. Journal of Abnormal Psychology,
108, 106–119.
Chassin, L., Presson, C., Pitts, S., & Sherman, S. J. (2000). The natural
history of cigarette smoking from adolescence to adulthood in a mid-
western community sample: Multiple trajectories and their psychosocial
correlates. Health Psychology, 19, 223–231.
Chassin, L., Rogosch, F., & Barrera, M. (1991). Substance use and symp-
tomatology among adolescent children of alcoholics. Journal of Abnor-
mal Psychology, 100, 449–463.
Chen, K., & Kandel, D. B. (1995). The natural history of drug use from
adolescence to the mid-thirties in a general population sample. American
Journal of Public Health, 85, 41–47.
Cloninger, C. R. (1987). Neurogenetic adaptive mechanisms in alcoholism.
Science, 236, 410–416.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences
(2nd ed.). Hillsdale, NJ: Erlbaum.
Colder, C. R., & Chassin, L. (1999). The psychosocial characteristics of
alcohol users versus problem users: Data from a study of adolescents at
risk. Development and Psychopathology, 11, 321–348.
Costello, E. J., Erkanli, A., Federman, E., & Angold, A. (1999). Develop-
ment of psychiatric comorbidity with substance abuse in adolescents:
Effects of timing and sex. Journal of Clinical Child Psychology, 28,
298–311.
Curran, P., Stice, E., & Chassin, L. (1997). The relation between adolescent
alcohol use and peer alcohol use: A longitudinal random coefficients
model. Journal of Consulting and Clinical Psychology, 65, 135–140.
Dishion, T., Patterson, G., & Reid, J. (1988). Parent and peer factors
associated with drug sampling in early adolescence. In E. R. Rahdert &
J. Grabowski (Eds.), Adolescent drug abuse: Analyses of treatment
research (NIDA Research Monograph 77, pp. 69–93). Rockville, MD:
National Institute on Drug Abuse.
Endicott, J., Andreason, N., & Spitzer, R. L. (1975). Family history
diagnostic criteria. New York: Biometrics Research, New York Psychi-
atric Institute.
Furman, W., & Buhrmester, D. (1985). Children’s perceptions of the
personal relationships in their social networks. Developmental Psychol-
ogy, 21, 1016–1024.
Hawkins, J. D., Catalano, R. F., & Miller, J. Y. (1992). Risk and protective
factors for alcohol and other drug problems in adolescence and early
adulthood: Implications for substance abuse prevention. Psychological
Bulletin, 112, 64–105.
Helzer, J., & Pryzbeck, T. (1988). The co-occurrence of alcoholism with
other psychiatric disorders in the general population and its impact on
treatment. Journal of Studies on Alcohol, 49, 219–224.
Hill, K. G., White, H. R., Chung, I. J., Hawkins, J. D., & Catalano, R. F.
(2000). Early adult outcomes of adolescent binge drinking: Person- and
variable-centered analyses of binge drinking trajectories. Alcoholism:
Clinical and Experimental Research, 24, 892–901.
Hill, S., Shen, S., Lowers, L., & Locke, J. (2000). Factors predicting the
onset of adolescent drinking in families at high risk for developing
alcoholism. Biological Psychiatry, 48, 265–275.
Hussong, A., & Chassin, L. (1993). The stress-negative affect model of
adolescent alcohol use: Disaggregating negative affect. Journal of Stud-
ies on Alcohol, 55, 707–718.
Hussong, A., Curran, P., & Chassin, L. (1998). Pathways of risk for
accelerated heavy alcohol use among adolescent children of alcoholics.
Journal of Abnormal Child Psychology, 26, 453–466.
Jessor, R., Donovan, J., & Costa, F. (1991). Beyond adolescence: Problem
behavior and young adult development. Cambridge, England: Cam-
bridge University Press.
Jessor, R., & Jessor, S. K. (1977). Problem behavior and psychosocial
development: A longitudinal study of youth. New York: Academic Press.
Johnston, L. D., O’Malley, P., & Bachman, J. (1999). National survey
results on drug use from the Monitoring the Future Study, 1975–1998:
Volume 1. Secondary students (USDHHS, National Institute on Drug
Abuse, Publication No. 99–4660). Washington, DC: U.S. Government
Printing Office.
Jones, B., Nagin, D., & Roeder, K. (1999). An SAS program based on
mixture models for estimating developmental trajectories [computer
software and manual]. Retrieved from http://www.stat.cmu.edu./
⬃bjones/traj.html
Kandel, D. B., & Chen, K. (2000). Types of marijuana users by longitu-
dinal course. Journal of Studies on Alcohol, 61, 367–378.
Labouvie, E., Pandina, R., White, H., & Johnson, V. (1990). Risk factors
of adolescent drug use: An affect-based interpretation. Journal of Sub-
stance Abuse, 2, 265–285.
Little, R., & Rubin, D. (1987). Statistical analysis with missing data. New
York: Wiley.
Merikangas, K. R. & Avenevoli, S. (2000). Implications of genetic epide-
miology for the prevention of substance use disorders. Addictive Behav-
iors, 25, 807–820.
Milich, R., Lynam, D., Zimmerman, R., Logan, T., Martin, C., Leukefeld,
C., et al. (2000). Differences in young adult psychopathology among
drug abstainers, experimenters, and frequent users. Journal of Substance
Abuse, 11, 69–80.
Muthen, B., & Shedden, K. (1999). Finite mixture modeling with mixture
outcomes using the EM algorithm. Biometrics, 55, 463–469.
Nagin, D. (1999). Analyzing developmental trajectories: A semi-
parametric group-based approach. Psychological Methods, 4, 139–157.
Robins, L., Helzer, H., Croughan, J., & Ratcliff, K. (1981). National
Institute of Mental Health Diagnostic Interview Schedule: Its history,
characteristics, and validity. Archives of General Psychiatry, 38,
381–389.
Rohde, P., Lewinsohn, P. M., & Seeley, J. R. (1996). Psychiatric comor-
bidity with problematic alcohol use in high school students. Journal of
the American Academy of Child and Adolescent Psychiatry, 35,
101–109.
Sandler, I. N., Ramirez, R., & Reynolds, K. (1986, August). Life stress for
77
SPECIAL SECTION: BINGE DRINKING TRAJECTORIES
children of divorce, bereaved, and asthmatic families. Paper presented at
the 94th Annual Convention of the American Psychological Association,
Washington, DC.
Schaefer, E. (1965). Children’s reports of parental behavior: An inventory.
Child Development, 36, 413–424.
Schuckit, M. (1994). Low level of response to alcohol as a predictor of
future alcoholism. American Journal of Psychiatry, 151, 184–189.
Schulenberg, J., Wadsworth, K., O’Malley, P., Bachman, J., & Johnston, L.
(1996). Adolescent risk factors for binge drinking during the transition
to young adulthood: Variable- and pattern-centered approaches to
change. Developmental Psychology, 32, 659–674.
Shedler, J., & Block, J. (1990). Adolescent drug use and psychological
health: A longitudinal inquiry. American Psychologist, 45, 612–630.
Sher, K. J. (1991). Children of alcoholics: A critical appraisal of theory
and research. Chicago: University of Chicago Press.
Sher, K. J., & Gotham, H. (1999). Pathological alcohol involvement: A
developmental disorder of young adulthood. Development and Psycho-
pathology, 11, 933–956.
Swaim, R., Oetting, E., Edwards, R., & Beauvais, F. (1989). Links from
emotional distress to adolescent drug use: A path model. Journal of
Consulting and Clinical Psychology, 57, 227–231.
Swendsen, J., & Merikangas, K. (2000). The cormorbidity of depression
and substance use disorders. Clinical Psychology Review, 20, 207–234.
Tarter, R., & Vanyukov, M. (1994). Alcoholism: A developmental disor-
der. Journal of Consulting and Clinical Psychology, 62, 1096–2007.
Tatsuoka, M. (1988). Multivariate statistics: Techniques for educational
and psychological research (2nd ed.). New York: Macmillan.
Weingardt, K. R., Baer, J. S., Kivlahan, D. R., Roberts, L., Miller, E., &
Marlatt, G. A. (1998). Episodic heavy drinking among college students:
Methodological issues and longitudinal perspectives. Psychology of Ad-
dictive Behaviors, 12, 155–167.
Wills, T. A., McNamara, G., Vaccaro, D., & Hirky, E. (1996). Escalating
substance use: A longitudinal grouping analysis from early to middle
adolescence. Journal of Abnormal Psychology, 105, 166–180.
Wills, T. A., Vaccaro, D., & McNamara, G. (1992). The role of life events,
family support, and competence in adolescent substance use: A test of
vulnerability and protective factors. American Journal of Community
Psychology, 20, 349–374.
Zucker, R., Boyd, G., & Howard, J. (1994). The development of alcohol
problems: Exploring the biospsychosocial matrix of risk (NIAAA Re-
search Monograph 26; NIH Publication No. 94–3495). Washington,
DC: U.S. Government Printing Service.
Received September 16, 2000
Revision received June 20, 2001
Accepted July 3, 2001 䡲
78
CHASSIN, PITTS, AND PROST
