A Comparison of Mechanisms Underlying SubstanceUse
for EarlyAdolescent
Childrenof Alcoholics
andControls*
BROOKES.G. MOLINA, PH.D.* LAURIECHASSIN PH.D. ANDPATRICKJ. CURRAN,M.A.
Departmentof Psychology,
ArizonaStateUniversity,Tempe,Arizona 85287
evidenceof differential importancefor COAs and non-COAs of the
parent monitoringand negative affect mechanisms.Parentalsocialization and negativeaffect mechanismssignificantlypredictedadolescent substanceuse regardlessof COA status. Differences did
emergeregardingthe effects of age and parent educationon peer
substanceuse and the effect of sociabilityon adolescentsubstance
use. (J. Stud. Alcohol 55: 269-275, 1994)
ABSTRACT. The currentstudyexamineddifferencesbetweenchildren of alcoholics(COAs) and controlsin parentmonitoring,stress-
negative affect, and temperamentmechanismsunderlyingearly
adolescentsubstanceuse. Using structuralequationmodeling, we
tested whether these mechanismswere equally predictiveof substanceusefor both groups.We extendedan earlier studythat tested
mediators of COA risk for substance use but did not examine COA
status as a moderator of these mechanisms. Overall, we found no
INDINGS
thatchildren
of alcoholics
(COAs)
areat
increased risk for adult alcoholism has stimulated in-
terest in uncoveringthe mechanismsunderlyingthis vulnerability (see Sher, 1991). Little is known, however,
aboutalcoholand drug useamongCOAs duringearly adolescencewhich is a period of risk for substanceuse initiation (Johnstonet al., 1988). Research has suggested
that adolescentsubstanceuse may result from multiple
mechanisms
includingpeermodelingand socialinfluence,
deficitsin parentalsocialization,controland support,parent modelingand toleranceof use, and attemptsto cope
with stress-induced
negativeaffect (seeChassin,1984, for
a review). However, it is unknown if the mechanisms
underlyingsubstance
use differ for high-riskadolescents
(suchas COAs) comparedto the generaladolescentpopulation. If particularmechanisms
were associatedwith substanceuse in a high-riskgroup,they would be of special
interest for preventive interventionefforts. The current
studytestedwhethermechanisms
underlyingearly adolescent substanceuse differed for a sample of COAs and
their non-COA peers.
Multiple pathwaysto substance
use have been posited
to accountfor the increasedrisk for substanceuse among
COAs. These mechanisms have included differential
sen-
sitivity to the positiveor negativeeffectsof alcohol,parent modelingof alcohol abuseand exposureto alcohol,
and use of alcoholand drugsto cope with tendenciesto-
Received:August4, 1992. Revision:February16, 1993.
*This studywas supportedby National Instituteon Drug Abusegrant
DA05227 to Laurie Chassin and Manuel Barrera, Jr.
*BrookeS.G. Molina is now with the WesternPsychiatric
Institute
andClinic, Universityof Pittsburgh,3811O'Hara Street,Pittsburgh,Pa.
15213-2593.Correspondence
shouldbe sentto her at this address.
ward negativeaffective states(see Sher, 1991). Recently,
Chassinand colleaguesfoundthat parentalalcoholismaffected early adolescentsubstanceuse throughstressand
negative affect mechanismsand through impairmentsin
parentmonitoring,both of which increasedthe probability
of associationswith a peer network that supportedsubstance use behavior. Parent alcoholism
stance use for
examine
COAs
and controls.
the role of COA
Such
a test would
status as moderator
of these
mechanisms.The current study extendedChassin et al.'s
work to addressthis issue:namely, are the processesthat
place adolescentsat risk for substanceuse similar for
COAs and non-COAs?The stress-negativeaffect mechanism was of specialinterest.Researchsuggeststhat young
adult COAs derive greaterstressresponsedampeningbenefits of alcohol use than do their non-COA peers (Levensonet al., 1987). Consequently,COAs may be more likely
to use substances
to cope with stress-induced
negativeaffective states.This hypothesissuggeststhat mechanisms
involvingstressand negativeaffect may be more important to substanceuse among COAs than among nonCOAs. To test this prediction, we used the same sample
and measuresreportedin Chassinet al. (1993) to compare
parameterestimatesfor the model in Figure 1 for COAs
and non-COAs.
269
was also associated
with higher levels of temperamentalemotionalityin adolescentswhich raised risk for experiencingnegative affect. These findings were producedusing a community
sample in which parent alcoholismwas directly ascertained and data were gatheredfrom multiple informants
(Chassin et al., 1993).
However,Chassinet al.'s studydid not include a test of
whether the parent monitoring, negative affect and temperament mechanismswere equally predictive of sub-
270
JOURNAL
OF STUDIES
ON ALCOHOL
/ MAY
1994
Father's
Monitoring
Mother's
Monitoring
I Child's I
.
.
] .26(.26L
'"-I Negative•
Child's
.34(.27) .,.•l
Stress
AffiliationsI
Child's
I .59(.6_0)
withDrug-•
Substance
I Affect
[•,. Using
Peers
[•.
Use
d3
Child's
Emotionality
I
/
•.14
(.20)
•
Child's
Sociability
FIGURE1. The hypothesized
structuralmodelof mechanisms
underlyingearly adolescent
substance
use. For presentational
ease,the exogenous
variablesof age, gender,parenteducation,currentparentalcoholconsumption,
andtheir associated
pathsare not shown.Pathcoefficientestimatesare
standardized
within eachgroup;estimatesin parentheses
are for controlsandwithoutparentheses
are for COAs. d l to d3 = structuraldisturbances.
Solid
linesindicatesignificanteffectsat p < .05 for bothgroups.Dashedlinesindicatemarginallysignificanteffectsfor bothgroups.(aThepathcoefficientfor
sociability-•substance
use is significantfor COAs [b = . 16, p < .05] but not for controls[b = -.04, NS].)
Method
Subjects
Of a total sampleof 454 adolescentsand their parents,
a subsampleof 327 familieswith completedata (provided
by two parentsand the adolescent)was usedfor the anal-
ysesreportedin the currentstudy.
I Adolescents
in this
subsampleranged in age from 10.5 to 15.5 years (mean
age = 12.7 years) and approximatelyhalf were female
(n = 176). In addition, 178 adolescentshad at least one
biologicalalcoholicparentwho was also a custodialparent (COAs), and the remaining 149 adolescentswere demographically matched controls with biological and
custodialparentshavingno historyof alcoholism.(For additional information, see Chassin et al., 1993.)
Recruitment
Recruitment proceduresare presentedin detail elsewhere (Chassin et al., 1991, 1992). COA families were
recruitedusing court records(full sample= 103), well-
nessquestionnaires
from a health maintenanceorganization (full sample= 22) andcommunitytelephonesurveys
(full sample= 120). (One family was referredby a local
VA hospital.)Demographicallymatchedcontrols(matched
on ethnicity,family composition,adolescentage within 1
year and SES) with no historyof parentalcoholismwere
recruitedover the telephoneusing reversedirectoriesto
identifyfamiliesin the sameneighborhoods
as COAs. Adolescentshad to be Hispanic or non-HispanicCaucasian
and 10.5-15.5 years old, English-speaking,and without
cognitivelimitationsthat would precludeinterview.Analysesto detect participationbias determinedthat magnitude of bias was small and unrelated to archival indicators
ofalcoholism
(seeChassin
etal., 1991,1992).
Procedure
Familieswere invited to participatein a studyof adolescent developmentand substanceuse. Data were col-
lectedby trainedinterviewersusingindividualcomputerassistedinterviewswith the adolescents
and their parents.
Interviewerswere blind to alcoholismstatusof the family.
MOLINA,
CHASSIN
AND CURRAN
271
To minimize contamination,family memberswere inter
viewed individually on one occasion by different interviewers; privacy was emphasized. Confidentiality
Johnstonet al., 1988; •t = .92) and of how close friends
was assured and reinforced
would feel about the adolescent's use of these substances
with
a DHHS
Certificate
of
Confidentiality.
both occasionallyand regularly (seven items; ot = .93).
The two scale scoreswere standardizedand averagedto
representthe proximal peer environment.
Measures
The measuresof interest were drawn from a larger
interview battery. (A completereport appearsin Chassin
et al., 1993.) All variableswere incorporatedinto the hypothesized structural model as manifest indicators and
constructionof these variables is describedbelow.2
Demographiccontrol variables. Child age, gender and
averagelevel of parent educationalattainmentwere used
as backgroundcontrol variables in the model.
Parent alcoholism.Lifetime DSM-III diagnosesof alcoholism (abuseor dependence)were obtainedfrom parent
interview usinga computerizedversionof the Diagnostic
Interview Schedule (DIS) (Robins et al., 1981).
Predictor variables in the hypothesized structural
model. Mother and father monitoring of child behavior
was assessedby parent self-report(3 items, e.g., "I had a
pretty good idea of [the child's] plansfor the day," Cronbach's •mother: .79, Cronbach'stlfather: .75). Child's
current life stresswas how many of 22 negative, uncontrollable
events
marijuanaandotherdrugsbothoccasionally
andregularly
(six itemsadaptedfrom the Monitoringthe Futurestudy;
had occurred
in the last
3 months
as
reportedby parentsand adolescents.
Adolescentsalso reported on four additionalpeer-relatedevents.Items were
taken from the Children
of Alcoholics
Life Events Sched-
ule (Roosa et al., 1988) and the General Life Events
Schedulefor Children (Sandleret al., 1986) supplemented
with several items from other children's
life events sched-
ules. For the structural model, the life stress variable was
a multiple reportercompositemanifestvariable usingfactor scoreregressionweights.Child emotionalityand sociability were measured using a modification of the adult
version of the Emotionality-Activity-SociabilityScale
(EAS) (Bussand Plomin, 1984). Coefficientalphasfor the
child, mother and father reports were, respectively,.72,
.78 and .76 for emotionality (8 items) and .46, .62 and
.60 for sociability (4 items). For the structural model,
these two temperamentvariables were multiple reporter
compositemanifestvariablescreatedusingfactorscoreregression weights. Child's negative affect was measured
using child self-report of internalizing symptomatology
(seven items from the Achenbach Child Behavior Checklist; Achenbach and Edelbrock, 1981; ot = .78), selfderogation (seven items from Rosenberg's 1977 scale;
ot = .81), and perceivedlossof control(three items from
Newcomb and Harlow, 1986; ot -- .73). For the structural
modeling, child's negative affect was a compositemanifest variablecreatedusingfactorscoreregressionweights.
Affiliation with drug-usingpeerswas measuredusing the
adolescent'sestimationof how many friends usedalcohol,
The dependentmeasure:Adolescentsubstanceuse. Adolescentsself-reportedtheir frequencyof consumptionof
beer/wine and distilled spirits (2 items), five or more
drinks in a row (1 item), getting drunk on alcohol (1
item), and eight illicit drugs(8 items) all in the past 3
months.Responseoptionsrangedfrom (0) not at all to (7)
every day. A singlesubstance
use scorewas calculatedby
summingthe responsesto the 12 items. A normalizing
transformationwas used to reduce skewness(after trans-
forming,skewness
was1.75).3
Results
Model specificationand adequacyof fit for the
full sample
Beforetestingfor differencesin pathestimatesbetween
COAs and controls,the fit of the hypothesizedmodel in
Figure1 wasestimated
forthefull sample
(n = 327).4 To
account for covariation among the backgroundcontrol
variables and the endogenousvariables, the model was
first estimatedwith all pathsfrom child age, genderand
parent educationfixed to zero; control paths with high
modificationindices(-> 5) were successivelyfreed until
noneremained.Only two pathswere significant:child age
to child negativeaffect and child age to affiliations with
drug-using peers. These two paths were included in the
structural
model.
The hypothesized,
structuralmodelfit the datavery well
(X2 = 18.75,15df, p > .22, TLI = .98, BBI = .98).5
Moreover, examination of the residuals and modification
indices further indicated that the observed variance/cova-
riance matrix was adequatelyreproducedby the model in
Figure 1. As describedin Chassinet al. (1993), lessparental monitoring was associatedwith greater adolescentaf-
filiationwithdrug-using
peers(bfather,
smonitoring
: --' 17,
p < .01; binother,
s monitoring
= --.09, p<. 13) and with
adolescentsubstance
use (blather.
s monitoring
= --'16,
p < .01); affiliation with drug-usingpeers was in turn
associated
with
increased
adolescent
substance
use
(b = .63, p < .01). Adolescentstresseventswere significantly related to adolescentnegative affect (b = .38,
p < .01) and negativeaffect was in turn significantlyrelated to affiliation with drug-using peers (b = .29,
p < .01). Negativeaffect was also marginallydirectly related to adolescentsubstanceuse (b = . 11, p < . 10). Adolescentemotionalitywas associatedwith higherlevelsof
adolescentnegativeaffect (b = .18, p < .05), while ad-
272
JOURNAL
OF STUDIES
ON ALCOHOL
TABLE1. Comparisons
of modelparameters
betweenCOAsandcontrols
Model
X2 (df)
TLI BBI
7•it-t(df)
131.71'(81)
.91 .85
-
51.44(42)
.97 .94 80.27*(39)
46.86 (41)
.98
.95
4.58* (l)
45.07 (40)
.98
.95
1.79 (1)
40.30 (39)
1.00
.95
4.77* (1)
39.02 (38)
1.00
.96
1.28 (1)
34.94 (37)
1.00
.96
4.18' (1)
3. Age-->peersfree within
groups
4. Parenteducation-->peers
added to model, invariant
betweengroups
5. Parent education-->peers
free within groups
6. Sociability-->substance
use added to model,
invariant between groups
7. Sociability-->substance
use free within groups
Notes: "Invariant" refers to specifiedparameter(s)estimatedas equal
for COAs and controls."Free within groups"refersto specifiedparameter(s) estimatedas not equal for COAs and controls.Path coefficient
parameterestimatesare denotedby arrowssurrounded
by the relevant
constructs(e.g., the effect of child age on affiliationswith drug-using
peersis denotedby age-->peers).
Peers= affiliationswith drug-using
peers.X2a,
,- indicates
testof significant
changein modelfit fromthe
previousmodel.
*p < .05.
becauseof previouslyestablishedlinks betweenparent alcoholism and environmental stress; Roosa et al., 1988).
1 for Model
2.
The hypothesizeddifferencesbetween COAs and controls with regard to the relation between negative affect
2. Variances/covariances
for exogenousvariables,
disturbances
for endogenous
variables,free within groups
1994
Model fit improvedsignificantlyand is displayedin Table
1. All parameterestimates
invariant
/ MAY
*p < .001.
and substance use were not found.
withdrug-using
peers(b = . 19,p < .01).6
indices
use were added to the model and released from invariance
(Models 5 and 7, respectively).Final path coefficientsfor
both groupsare in Figure 1. Child age was more strongly
relatedto affiliationswith drug-usingpeersfor COAs than
for controls,parenteducationwas more stronglyrelatedto
affiliations with drug-usingpeers for controlsthan for
COAs, and sociabilitywas more stronglyrelatedto child
substance use for COAs
olescentsociabilitywas associatedwith less negative affect (b = -.28, p < .01) but also with greateraffiliation
Modification
associatedwith thesetwo paths (negative affect to affiliations with drug-usingpeers, negative affect to substance
use) were very low (<- 1) both beforeand after freeing the
parametersdescribedabove, indicatinga lack of significant differencesin the paths from negative affect for the
two groups.However, severaldifferencesbetweengroups
were identified by high modificationindices.In Table I it
can be seen that there was significant improvementin
modelfit after the path from child age to affiliationswith
drug-usingpeerswas releasedfrom invariance(Model 3)
and after the paths from parent educationto affiliations
with drug-usingpeers and from sociability to substance
than for controls.
There were no
other indications of reliably significant differencesbetween groups in any of the remaining path coefficient
estimates.
? FinalmultipleR2 estimates
for childnegative
affect, affiliations with drug-usingpeers and child sub-
Model fit for COAsand controls
stanceuse were .36, .38, .54 for COAs, and .43, .32, .51
An iterative series of cross-groupanalyseswas conductedto determinethe presenceof significantdifferences
for controls, respectively.These estimateseach differed
significantlybetweengroupsat p<.05 as indicatedby significantincreasesin the modelchi squarewheneachesti-
between COAs
matewasindividually
specified
asinvariant
(•2diff
, 1 df,
and controls in overall model fit and in
individualparameterestimates.First, the overallfit of the
structuralmodel in Figure I was comparedfor COAs and
controls.A cross-groupcomparisonwas conductedwith
all parameter estimates specified as invariant across
groups.The resultingfit, displayedin Table I for Model
1, was less than adequateand confirmedthe presenceof
significantdifferencesin overall model fit.
ranged from 4.60 to 6.30, all p<.05). Overall, despite
thesedifferencesin predictedvariance, the findings suggestedthat the parent monitoringand stressand negative
affect mechanismswere similarly related to substanceuse
for COAs and controls.a
Discussion
Second, differences between COAs and controls in indi-
vidual parameterestimateswere examined.The presence
of multiple modificationindicesgreaterthan five indicated that the variance/covariancematrix for the exogenous variables
and the disturbances associated with the
endogenousvariables were significantlydifferent across
groups.Theseparameters
werefreed(1) to improvemodel
fit for both groupsin order to have increasedconfidence
in path coefficientdifferencesand (2) becausedifferences
in the variance/covariance
and disturbanceparameterestimatesacrossgroupswere consistentwith theory-basedexpectations(e.g., greatervariabilityin stressamongCOAs
The goal of this study was to considerpossibledifferences between COAs
and non-COAs
in the mechanisms
underlyingearly adolescentsubstanceuse. In particular,
we hypothesizedthat negative affect would relate more
stronglyto substanceuse (both directly and indirectlyvia
involvement with substance-usingpeers) for COAs than
for non-COAs. Our findings did not supportthis prediction. Rather,our findingsindicatedthat parentmonitoring
and stressand negative affect mechanismswere significantly related to substanceuse for both COA and nonCOA adolescents,and the degreeof associationwas not
MOLINA,
CHASSIN
differentbetweenthe two groups.This finding is notewor-
thy in view of Chassinet al.'s (1993) finding that COA
parentsmonitoredtheir childrenless and COAs experienced significantlymore stressand negativeaffect than
controls. Thus, although early adolescentCOAs show
higherlevelsof substance
use and higherlevelsof these
particularpsychosocial
risk factors,thesemechanisms
underlying substance
use appearto operateboth for those
with and withoutalcoholicparents.Thesefindingssuggest
that programsaimed at teachingcopingskills or at improvingparentmanagement
couldbe usefulregardless
of
parentalcoholismhistory.
Our finding that negative affect mechanismsoperated
similarlyfor COAs and controlsmay be due to the early
agesof our adolescents.
We purposefullysampledyouth
in the 10-15year old age rangewith the aim of capturing
the period of risk for substanceuse initiation (Johnston
et al., 1988). This strategy resulted in our having few
substance-abusing
or substance-dependent
adolescents--
the populationtypicallydescribedwith regardto negative
affect regulationmotivesfor substanceuse. Particularly
strongnegative affect regulationmotives for use among
COAs (comparedto non-COAs) may not emerge until
older agesand/orhigherlevelsof use.
Evidencethat girls, more than boys, report drinking to
cope with stressors
(Windie, 1991) suggeststhat gender
ratherthan parentalcoholismmay moderatethe impactof
stress-negative
affect mechanisms
on substance
use. However, Chassinet al. (1993) did not find genderdifferences
in the stressand negativeaffect mechanismwhen testing
their mediationalmodel for boys and girls separately.A
moderatingeffect of gendermay not emergeuntil later in
AND CURRAN
273
that put COAs at risk for substanceuse. These interactions may include friends who are deviant in other ways
(not directly reflected in their substanceuse) or interactions with drug-usingindividualswho are not perceivedas
"friends" (e.g., older relative, neighbors).Alternatively,
sociabilitymay not only elevate associationswith deviant
peers but also may serve as a marker for a disinhibited
behavioralstyle that hasbeen associatedwith the development of alcoholism, with substanceuse and with greater
stressresponsedampeningeffects of alcohol (Sher and
Levenson, 1982; for reviews, see Tarter et al., 1985; Windie, 1990). These mechanisms deserve future research at-
tention in explaining the differential risk for adolescent
substanceuse associatedwith parent alcoholism.
Overall, our findings suggestedthat parent monitoring
and negative affect mechanismssignificantly predicted
early adolescentsubstanceuse regardlessof parental history of alcoholism.Our failure to find a predictedstronger
effect of the negative affect mechanismfor COAs persistedindependentlyof our considerationof currentparent
alcoholconsumption.However,differentdefinitionsof parental alcoholism,different agesof COAs or different operationalizationsof constructsmight all affect findings
(Sher, 1991), and our results are in need of replication
with other samplesand longitudinaldesigns.
Acknowledgments
The authorsthank Mark Reiserand StephenWest for statisticalconsultation and comments on this article.
Notes
adolescence.
Differences
between COAs and non-COAs
were found
for the effectsof child age and parenteducationon affiliations with drug-usingpeers, and for the effect of sociability on child susbtanceuse. Age was more strongly
relatedto affiliationswith drug-usingpeersfor COAs than
for controls.In other words, as they approachmiddle adolescence,COAs may escalatemore rapidly into a highrisk environmentmarked by relatively greater immersion
in a drug-tolerantpeer network. Furthermore,for COAs,
theseassociations
were independent
of socioeconomic
status (as defined by parent education). For controls, however, a drug-tolerantpeer environmentwas related to
parent education, suggestingthat broader sociocultural
factorsaffect peer choice in the absenceof a family history of alcoholism.
We expectedsociability to exert all of its influence
throughaffiliationswith drug-usingpeers.Thus, our finding that sociabilitywas directly (and more strongly)related to substance use for COAs than for non-COAs
was
surprising.Perhapsour operationalizationof peer influence as friends who use drugs or who tolerate drug use
was too narrow to captureall of the social interactions
When investigatingthe mechanismsunderlyingthe relation between
COA statusand adolescentsubstanceuse, Chassin et al. (1993) tested
their hypothesized
mediationalmodelusingthe subsample
of families
with complete data (n = 327) and the entire sample of two-parent
families(n = 416) wheredata for one noninterviewedparentwas imputed (BMDP AM program, Version 5; see also Little and Rubin,
1987). Similar findings were producedusing the samplesof 327 and
416 suchthat, with only one exception,all pathsreportedas significant usingthe sampleof 327 remainedsignificantusingthe sampleof
416. The one changewas that the effects of emotionalitybypassed
negativeaffect to predictassociations
with drug-usingpeers.In addition, severalpathsthat were nonsignificantor marginally significant
in the original model became significant using the imputed data:
mother'salcoholismwas significantlyassociatedwith lower levelsof
maternalmonitoring,lower maternalmonitoringwas associatedwith
higher levels of peer drug use and father's alcoholismwas significantlyassociated
with adolescent's
heightenedemotionality.Thus, reestimatingthe modelusingimputeddata for the total sampleof twoparentfamiliescloselyreplicatedthe findingsbasedon the sampleof
327 two-parentfamilies with completedata.
In addition, comparisonswere performedbetweentwo-parentand
single-parentfamilies (t testsand chi squares)to assesswhether the
exclusionof single-parentfamilies (33 single mothersand 5 single
fathers)introducedbias into the sample.The groupswere compared
on a variety of demographic
variablesas well as on all variablesrepresentedin the hypothesizedmediationalmodel tested by Chassin
274
JOURNAL
OF STUDIES
ON ALCOHOL
et al. (1993). There were no significant differencesbetween the
single-parent
and two-parentfamiliesfor 18 of the 29 comparisons
including, most importantly,the adolescent'salcohol- or drug-use
outcomes.Of the significantdifferencesthat were found, the directions of the differenceswere not consistentin terms of puttingeither
the single-parentor the two-parentfamily at higher risk. Moreover,
the differencesbetweenthe two groupswere generallyminor.
2. Compositemanifestvariablesfor constructsinvolvingmultiple reporters(stress,emotionality,sociability)or multipleindicators(negative affect) were createdto preserveinformation from the multiple
reportersand multiple indicators.The inclusionof theseas latent
variableswas untenablegiven the requirednumberof parametersand
current sample size which would have exceededBollen's (1989,
p. 268) suggestion
to haveat leastseveralcasesper free parameter.
To createthe multiplereportervariables,weightedlinearcomposites
of mother,fatherandchild reportof stress,emotionalityand sociability were calculatedusingfactorscoreregressionweightstaken from a
singlethree-factormeasurement
modelof theseconstructs.
The negative affect variable was constructedsimilarly using weights taken
from a single one-factormeasurementmodel (see Loehlin, 1987, for
a more detaileddiscussionof factor-scoreregressions).Compositereliabilities were calculatedfor each manifestvariable using the estimatesof error and total varianceproducedby the measurement
model
(Lord and Novick, 1968, p. 86). The error terms for thesemanifest
variables,and for affiliationswith drug-usingpeersand child's substanceuse, were then set to one minus the reliability multiplied by
the varianceof the indicator.(For additional details concerningthese
/ MAY
1994
exogenousvariable affectingchild's substanceuse. This singlemanifest variablewas not includedin the original model testedabovebecause(a) Chassinet al. (1993) determinedthat effectsof parental
alcohol consumptionwere different for mothers and fathers, and
(b) becauseinclusionof both motherand fatheralcoholconsumption
in the current article was untenabledue to the requirednumberof
parametersandsamplesize. However,in an effort to approximatethe
effectof currentparentdrinking,a singlemanifestvariablewas createdusingparentself-reportof alcoholquantityand frequencyof use
in the past 3 months.Quantity-frequency
productswere calculated
separatelyfor eachparentand naturallog transformations
were used
to reduce skewness below 1.00 as in Chassin et al., 1993. Of the two
parentreports,the one reportinghigher quantity-frequency
was chosenin orderto reflect currentparentalalcoholconsumption.
The predicted paths of theoreticalimportance(i.e., in Figure 1) were all
replicated, with parent monitoringand stressand negative affect
mechanismssignificantlyrelatedto substanceuse for both COA and
non-COA adolescents.
Current parentdrinking was significantlyrelatedto child'ssubstance
usefor both COAs (b = .14, p • .01) and
controls(b = . 13, p • .01), and all of the groupdifferencesin path
coefficients
wereconfirmed.
FinalmultipleR2 estimates
were:child's
negativeaffect, .36; affiliationswith drug-usingpeers, .38; child's
substance
use, .56 for COAs; and .43, .32 and .54, respectively,for
controls;and these estimatesdiffered significantlybetweengroups
(p • .05). The final model reproducedthe observedvariance/covari-
ancematrixvery well (•2 = 40.07, 42 dr, p > .55, BBI = .96,
TLI = 1.00).
measures see Chassin et al., 1993).
3. The normalizingtransformation
was conductedfor the full sampleof
327 to retain a common
metric
for the substance use variable
COAs and controls--a requirementfor comparingcovariancestructuresacrossgroups(Joreskogand Sorbom, 1989).
4. All analyseswere basedon the variance/covariance
matrix in LISREL
VII under maximum
likelihood
estimation.
5. The Tucker-Lewis (TLI) and Bentler-Bonnett(BBI) indices of incre-
mentalfit rangefrom 0 to 1.00; Bentierand Bonnett(1980) recommend .90 as a minimal criterion for acceptablemodel fit.
6. Parameterestimates(b's) are standardized.Whenpresentedfor crossgroup analyses,b's are taken from LISREL'S
within-groupstandardized solution.
7. One final modification index equal to 6.22 remained for the path
from emotionalityto negativeaffect. When this path was freely estimatedwithin groups,emotionalitywas significantlyrelatedto negative affect for controls(b = .34, p < .01) but not COAs (b = .09,
NS) and model fit improvedsignificantly(•2a,rr= 6.52, I df,
p < .05). However,the estimatedcorrelationsbetweenemotionality
and stressexceeded1.0 for both groupsand differedslightlyin magnitudebetweenCOAs and controls,promptingconcernthat the path
coefficientdifferencewasunstableanddueto differentialcollinearity
betweengroups.To test the possibilitythat over-correctionfor measurementerror causedthe out-of-boundcorrelationand, subsequently,
the group differencesin the path coefficientestimates,the final
cross-group
modelwas testedwith the measurement
errorsfor stress
andemotionality
setto zero(X2 = 31.49, 37 dr, p > .72). Statistically significantrelationsbetweenstressand negativeaffect and between emotionalityand negativeaffect were found for both groups,
but the modificationindicesno longersuggested
a differencebetween
groups in the relation between emotionality and negative affect.
Therefore,althoughthe relationsamongstress,emotionalityandnegative affect were reliably significantboth with and without correction
for measurement error, the COA/non-COA
difference in the relation
betweenemotionalityandnegativeaffectwasdeterminedto be dueto
over-correction
for measurement
References
for
error and therefore unstable.
8. To determinewhetherour findingswere independent
of currentparent drinking, the seriesof cross-grouptestsconductedabovewas repeated with current parent alcohol consumptionas an additional
ACHENBACH,T.M. ANt) EDELBROCK,C. Behavioral Problems and Com-
petenciesReported by Parentsof Normal and DisturbedChildren
Aged Four throughSixteen, Chicago:Univ. of ChicagoPress, 1981.
BENTLER,P.M. AND BONNETT,D.G. Significancetestsand goodness
of fit in the analysisof covariancestructures.Psychol.Bull. 88: 588606, 1980.
BOLLEN,K.A. StructuralEquationswith Latent Variables, New York:
John Wiley & Sons, Inc_, 1989.
Buss,A.H. ANDPLOMIN,R. Temperament:
Early DevelopingPersonality Traits, Hillsdale, N.J.: LawrenceErlbaum Assocs.,Inc., 1984.
CHASSIN, L. Adolescent substance use and abuse. In: KAROLY, P. AND
STEFFEN,J.J. (Eds.) Advancesin Child Behavioral Analysis and
Therapy,Vol. 3, Lexington,Mass.:LexingtonBooks, 1984,pp. 99152.
CHASSIN, L., BARRERA,M., JR., BECH, K. AND KOSSAK-FULLER,J.
Recruitinga communitysampleof adolescentchildrenof alcoholics:
A comparison
of threesubjectsources.J. Stud. Alcohol53:316-319,
1992.
CHASS•N,L., PILLOW, D.R., CURRAN, P.J., MOLINA, B.S.G. AND BAR-
RERA,M., JR. The relationof parentalcoholismto early adolescent
substanceuse: A test of three mediating mechanisms.J. Abnorm.
Psychol. 102: 3-19, 1993.
CHASSIN,L., ROGOsCH,
F. ANDBARRERA,
g. Substance
useand symptomatologyamongadolescentchildrenof alcoholics.J. Abnorm. Psychol. 100: 449-463, 1991.
JOHNSTON,L., O'MALLEY, P. AND BACHMAN,J. Illicit Drug Use,
Smoking,andDrinkingby America'sHigh SchoolStudents,College
Students,and YoungAdults, 1975-1987.Preparedfor the National
Institute on Drug Abuse, DHHS PublicationNo. (ADM) 89-1602,
Washington:GovernmentPrinting Office, 1988.
JORESKOG,
K.G. AND SORBOM,
D. LISREL 7: A Guide to the Program
and Applications,Chicago:SPSSInc., 1989.
LEVENSON, R.W.,
OYAMA, O.N. AND MEEK, P.S. Greater reinforce-
ment from alcoholfor thoseat risk: Parentalrisk, personalityrisk,
and sex. J. Abnorm. Psychol.96: 242-253, 1987.
MOLINA,
CHASSIN
LITTLE, R.J.A. AND RUBIN, D.B. Statistical Analysis with Missing
Data, New York: John Wiley & Sons, Inc., 1987.
LOEHLIN, J.C. Latent Variable Models: An Introduction to Factor, Path,
and StructuralAnalysis, Hillsdale, N.J.: LawrenceErlbaum Assocs.,
Inc., 1987.
LORD, F.M. AND NOVICK, M.R. Statistical Theories of Mental Test
Scores,Reading,Mass.:Addison-Wesley
PublishingCo., Inc., 1968.
NEWCOMB, M.D.
AND HARLOW, L.L. Life events and substance use
amongadolescents:
Mediatingeffectsof perceivedlossof controland
meaninglesshess
in life. J. Pets. Social Psychol.51: 564-577, 1986.
ROBINS, L.N.,
HELZER, J.E., CROUGHAN, J. AND RATCLIFF, K.S.
National Instituteof Mental Health DiagnosticInterview Schedule:
Its history,characteristics,
and validity. Arch. Gen. Psychiat.38:
381-389, 198I.
ROOSA,M.W., SANDLER,I.N., GEHRING, M., BEALS, J. AND CAPPO,
L. The children of alcoholics life-events schedule: A stress scale for
childrenof alcohol-abusing
parents.J. Stud. Alcohol 49: 422-429,
1988.
ROSENBERG,
M. Conceiving the Self, New York: Basic Books, Inc.,
1979.
AND CURRAN
275
SANDLER, I., RAMIREZ, R. AND REYNOLDS, K. Life stress for children
of divorce, bereaved,and asthmaticchildren. Paperpresentedat the
annualmeetingof the American PsychologicalAssociation,Washington, D.C., 1986.
SHER,K.J. Childrenof Alcoholics:A Critical Appraisalof Theory and
Research,Chicago:Univ. of ChicagoPress,1991.
SnER, K.J. AND LEVENSON, R.W.
Risk for alcoholism and individual
differencesin the stress-response-dampening
effectof alcohol.J. Abnorm. Psychol.91: 350-367, 1982.
TARTER,R., ALTERMAN,A. AND EDWARDS,
K. Vulnerabilityto alcoholismin men: A behavior-geneticperspective.J. Stud. Alcohol 46:
329-356, 1985.
WINDLE,M. Temperamentand personalityattributesof childrenof alcoholics. In: WINDLE, M. AND SEARLES,J.S. (Eds.) Children of Alco-
holics:Critical Perspectives,New York: Guilford Press, 1990, pp.
129-167.
WINDLE,M. Genderdifferencesin adolescentsubstance
use: A personenvironmentperspective.Paper presentedat the annual meeting of
the American PsychologicalAssociation, San Francisco, Calif.,
199 I.