Differential Influence of Parental Smoking and Friends' Smoking on Adolescent Initiation and Escalation and Smoking

Differential I&hence of Parental Smoking and Friends’Smoking on Adolescent Initiation and Escalation of Smoking*

BRIAN R. FLAYFRANK:B . HU _ ,_

_

OHIDUL SIDDIQUF.L . EDWhD D A Y

.

DONALD HEDEKERUniversi&of ll.!+ois at Chicago

JOHN PETRAITISUniversity of Alaska

JEAN RICHARDSONSTEVE SUSSMAN

University of Southern California

,

Journal of Health and Social Behavior 1994. Vol. 35 (September):248-265 - .

Smoking-related behaviors and attitudes of significant others (especial1~~friemi.s andparents) are among the most consistent predictors of adolescent smoking. However,theorists remain divided on whether the behaviors of significant others influenceadolescent smoking directly or indirectly, and the relative influence of parental andpeer smoking on adolescents’ own smoking is still a matter of debate. In addition,little research has examined the role of significant others’ behavior on different stagesof smoking onset. In particular, not much information is available regarding genderand ethnic diflerences in social influences on smoking behavior. We use structuralequation modeling to address these issues. Different theoretical perspectives fhontcognitive-affective theories (Ajzen 1985; Ajzen and Fishbein 1980) ana’ social learn-ing theories (Akers et al. 1979; Bandura 1969, 1982, 19861 have been integrated intoa structural model of smoking influence. The results show that friends’ smokingaffects adolescent initiation into smoking borh directly and indirectl$ whereas pa-rental smoking influences smoking initiation only indirectly. The data also show thatfriends’ and parents’ smoking affect smoking escalation only indirectly. In general,friends’ smoking has a stronger effect on adolescents’ smoking behavior, particularlyon initiation. Multiple group comparisons of the structural models predicting smokinginitiation among males and females reveal that parental approval of smoking plays asignificant mediating role for females, but not for males. Comparisons of Whites,Blacks, Hispanics, and other ethnic groups reveal that there are some significantdifferences in the pathways of friends ’ influences among the four groups.

INTRODUCTION

Numdrous studies have demonstrated thatadolescents whose friends and/or parents usesubstances are more likely to use substances

* Collection of data for this research was sup-ported by grant #ROl-DA0348 from the NationalInstitute of Drug Abuse to B.R. Flay. W.B. Hansen

and to use them more frequently than youthwhose friends and/or parents are nonusers

and C.A. Johnson. The analyses reported here werecompleted with support from grant #ROl-DA06307to B .R. Flay and J. Richardson. Address correspon-dence to Brian R. Flay, Prevention Research Center(M/C 275), University of Illinois at Chicago, 850West Jackson Boulevard, Suite 400, Chicago, Illi-nois 60607-3025.

S M O K I N G I N F L U E N C E

(Needle et al, 1986). Smoking by significantothers (especially friends) is one of the mostconsistent predictors of adolescent smoking.Most recently, Conrad, Flay, and Hill (1992)conducted a comprehensive-review of pro-spective studies on predicting onset ofcigarette smoking among adolescents. Of allstudies reviewed, 13 examined the influencesof parental smoking with seven showingsignificant results (Charlton and Blair 1989;Chassin et al. 1984, 1986; Goddard 1990;Murray et al. 1983; Semmer et al. 1987a;Semmer et al. 1987b). and 19 examinedfriends’ influences with 16 showing signifi-cant results (Alexander et al. 1983; Ary et al.1983; Ary and Biglan 1988; Bauman et al.1984; Charlton and Blair 1989 [females only];Chassin et al. 1984, 1986; de Vries et al.1990; McCaul et al. 1982; Mittelmark et al.1987; Murray et al. 1983; Semmer et al.1987a, 1987b; Stacy et al., unpublished;Sussman et al. 1987 [Asians only): Urberg,Cheng, and Shyu 1991).

pills, however, Newcomb, Huba, and Bentler .(1983) found an indirect effect for alcohol andpill use and a direct effect for marijuana use.In addition, Biddle, Bank, and Marlin (1980)found that, although parents and peers hadboth direct and indirect influences on adoles-cents’ use of alcohol, peers were more likelyto influence adolescents through direct mod-eling, while parents were more likely to exerttheir influence through norms.

Although the total effects have been welldocumented, considerable disagreement re-mains regarding how these effects operate.For example, it is unclear whether the effectsare direct or indirect, and the relativeinfluence of parental and friends’ smoking onadolescents’ smoking is still a matter ofdebate. In addition, research has seldomexamined the differential effects of parentaland friends’ behavior on different stages ofsmoking, and little is known about gender andethnic differences in the influences of parentsand friends on adolescents’ smoking. Thepresent research addresses these issues.

Direct or Indirect Effects

Several different theories have been pro-posed to explain how significant others’behaviors influence adolescent behavior. Theclassic modeling theory (Bandura 1969)suggests a direct effect. It postulates- thatparents or friends influence adolescent drugbehavior by setting an example and providingsocial reinforcement. The theory emphasizesthat direct observation and imitation of aparticular behavior of significant others arethe essential processes for adolescents toacquire that behavior. Akers’ social learningtheory (Akers et al. 1979:638) also empha-sizes that “social behavior is acquired throughdirect conditioning and through imitation ormodeling of others’ behavior.” By contrast,cognitive-affective theorists (Ajzen 1985;Ajzen and Fishbein 1980; Fishbein and Ajzen1975) have argued that the influences ofothers’ behavior operate indirectly and aremediated through cognitive-affective pro-cesses. In particular, Ajzen’s theory ofplanned behavior (TPB) (Ajzen 1985) insiststhat intentions are the immediate determinantof behavior. Intentions, in turn, are deter-mined by three constructs: attitudes towardthe behavior, subjective normative beliefs,and self-efficacy. TPB claims that otherfactors beyond the three constructs can onlyinfluence adolescents’ behavior by shapingthese constructs.

Although researchers agree that behaviors Other theoretical frameworks, however,of others are important predictors of adoles- have recognized the importance of both directcent behavior, they do not agree about how modeling and cognitive mediation. For in-others’ behaviors influence adolescents. For stance, Bandura’s social cognitive/learninginstance, investigators remain divided on theory has modified the classic modelingwhether others’ behaviors influence adoles- theory by including other mediating con-cents’ behavior directly or indirectly. Lau, structs, such as outcome expectations and..Quadrel, and Hartman (1990) investigated self-efficacy (Bandura 1982, 1986). Accord-influences of parents and peers on adolescent ing to the theory, observing role models whodrinking, diet, exercise, and wearing of seat use drugs will not only directly affectbelts and found that direct modeling of health adolescents’ own drug behavior, but will alsobehaviors was the most important channel of shape adolescents’ outcome expectations,influence from parents and friends. In a study which are their beliefs about the mostof the influence of mothers’ behaviors on immediate and most likely social, personal,their children’s use of alcohol, marijuana, and and physiological consequences of druo use.

.*

.:: . ~.~. ._*_i_i... :L.-..LAL-,. , . .,~u:. ,. . . _ . . _ . . : .&,’ -,.r .:. ,’ . . . . . :.. ,.

250 JOURNAL OF HEALTH Ah’D SOCIAL BEHAVIOR

In addition, role models can shape adoles-cents’ self-efficacy. For instance. observingpe@!rs purchase and inhale cigarettes canprovide adolescents with the necessaryknowledge and skills to obtain and usetobacco (use .- self-efficacy).- Conversely,knowing and observing a close friend usingdrugs may erode adolescents’ skills to refusedrug offers (refusal self-efficacy).

Relative Injluences of Parents and Friends

_’

-.

:

Another unresolved issue concerns therelative influence of parents and peers onadolescents’ behavior. It is agreed that bothparents’ and peers’ drug-using attitudes andbehaviors have a significant impact uponadolescent drug use, but the relative influ-ence of parents and peers seems to vary withthe use of different -drugs (Chassin et al.1986; Glynn 1981). Parents and peers arefound to be equivalently influential onadolescent alcohol use. While peer influ-ences predominate for marijuana use, paren-tal influences appear to play a moreimportant role in the use of illicit drugsother than marijuana. In addition, peerinfluences appear to be more influential inthe earlier phases of drug use (e.g., usingsoft drugs), whereas parental influencesbecome more important for advanced phasesof drug use (e.g., using hard drugs) andtend to be of long-term impact.

Many studies have demonstrated thatfriends’ smoking is a better predictor of anadolescent’s smoking than is parental smok-ing (e.g., Alexander et al.. 1983; Chassin etal. 1986; McCaul et al. 1982). One possibleexplanation comes from Kandel and An-drews (1987), who argue that the. strength ofinfluence from parents and friends dependson the quality of their social bond withadolescents. Given that adolescence is aperiod of increasing bonds with peers, andpossibly of weakening bonds with parents,smokin’g among friends is expected to exertmore influence on an adolescents’ ownsmoking. If this is so, increasing peerinfluence is more likely to produce a directpath between friends’ smoking and adoles-cents’ smoking than between parental smok-ing and adolescents’ smoking. However, thisnotion has never been tested with cigarettesmoking.

Stages of Cigarette Smoking

Adolescent smoking is a dynamic ratherthan a static phenomenon (Flay et al. 1983).The process of becoming a smoker has beensegmented into several stages, reflectingdevelopmental changes and the transitionsfrom never smoking to initiation of firstcigarette, and from initiation of smoking to ahigher level of smoking.. There is someevidence that the influences of parents andpeers may vary with different stages of druguse. Kandel and Andrews (1987) examinedthe processes of socialization by parents andpeers with respect to adolescent use of alc’oholand marijuana and found that modeling ofparental behavior played the most importantrole in the early process of initiation, whereasmodeling of peer behavior was more likely tooccur after the stage of experimentation. Inaddition, Skinner et al. (1985) found thatattachment to father and mother and associa-tion with female smoking friends were relatedto the initiation stage of cigarette smoking,while attachment to father and associationwith both male and female smoking friendspredicted cessation. There has been no studyattempting to examine possible differentprocesses of social influences at differentstages of cigarette smoking.

Gender Differences in Social Influences

There have been few systematic, focusedinvestigation;, of gender differences in paren-tal and peer influence on adolescent drug use.Existing literature suggests some genderdifferences in social influences. For instance,Chassin et al. (1986) found that peer andparent influences were significant for girls butnot for boys. In a study of 7th through 9thgrades in 11 schools in Los Angeles, Hubaand Bentler (1980) found that the perceivednumber of peer drug-use models was asignificant predictor of drug use for girls atall ages, but only at the high school level forboys. More recently, Waldron, Lye, andBrandon (199 1) found that the associationbetween smoking initiation and friends’smoking was stronger for girls than for boys.These findings suggest that girls may be moresusceptible to social influences than boys.Part of such gender differences might beexplained by females’ stronger social bondsto parents and schools than bovs (Ensminaer.

SMOKING INFLUENCE 251

Brown, and Kellam 1982). Another pbssibil-ity lies in the hypothesis that girls are morepeer-oriented than boys, which results inhigher levels of exposure to social influencesfor girls.. Previous research has found thatgirls tend to spend more time with friends ofthe opposite sex and are more involved insocial activities such as dancing and youthclubs (Swan, Creeser, and Murray 1990).

tion strategies-should be tailored to involveparents and friends by taking into account thesocial and ethnic background of the targetpopulation. The present research goes beyondprevious studies by dissecting the direct andindirect effects of parental and friends’influences on adolescents* smoking behaviorand, therefore, delineating the potentiallydifferent processes of social influences among ’different ethnic groups. .

Ethnic Differences in Social Influences

Ethnic@ has been considered as an impor-tant moderating factor for population differ-ences in social influences on adolescent druguse (Castro et al. 1987). In a study of the use offour types of drugs (beer/wine, hard liquor,marijuana, and pills) among Black, Hispanic,Asian, and White teenagers, Newcomb andBentler (1983) found different effects of peerand adult drug use on adolescents’ self-use fordifferent ethnic groups. In particular, for allthe substances, Blacks reported the leastimpact of peer use on self-use relative to otherethnic groups; for beer or wine, Blacks andHispanics reported the least impact of adultuse on self-use; for marijuana, Asians re-ported the lowest impact of adult use; forpills, Whites reported the highest impact ofadult use. Sussman et al. (1987) compared thepsychosocial predictors of smoking onsetamong 7th- and 8th-grade White, Black,Hispanic, and Asian students, and found thatsome unique variables predicted smokingonset in different ethnic groups. For Whites,adult and peer models of smoking behaviorwere among the strongest predictors; forBlacks, risk-taking preference seemed to bemore important than social influence vari-ables; for Hispanics, adult and peer approvalof smoking were among the most importantdeterminants of adolescents’ own smoking;for Asians, influences from peers and familyon cigarette smoking appeared to be secon-dary to the effects of poor school performanceand low self-esteem.

A Structural Model of Smoking Influence

Examination of the potentially differentmechanisms through which parental andfriends’ influences operate in different ethnicgroups may contribute to the understanding oftheoretical causes of cigarette smoking amongadolescents in different cultures. An under-standing of these differences, in turn, canprovide more specific guidance on howadolescent smoking orevention and interven-

By integrating different perspectives fromvarious theories described previously, wepropose a model of smoking influence forconducting a causal analysis of factorsassociated with adolescent smoking behaviors _such as initiation and escalation. The modelexamines the interactive effects of eightconstructs (Figure 1): (1) Friends’ Smoking,(2) Parental Smoking, (3) Negative OutcomeExpectation, (4) Perceived Friends’ Approvalof Smoking, (5) Perceived Parental Approvalof Smoking, (6) Refusal Self-efficacy, (7)Smoking Intentions, and (8) AdolescentSmoking Behavior, representing smokinginitiation and escalation, respectively, in themodels of smoking initiation and escalation.We hypothesize that exposure to smokingparents and friends directly triggers smokingbehavior (direct effects). We also hypothesizeindirect effects as follows: (1) exposure tosmoking parents-and friends is expected todiminish adolescents’ negative outcome ex-pectations about cigarette smoking and theirrefusal self-efficacy; (2) such exposure willfoster adolescents’ perception of approval ofsmoking by parents and friends; and (3)diminished negative outcome expectationsabout cigarette smoking, elevated perceivedapproval of smoking, and lower refusalself-efficacy, in turn, are expected to affectadolescents* smoking behavior through pro-moting adolescents’ smoking intentions. Wealso hypothesize that the disturbance terms ofsome of the med,iating constructs will becorrelated. Finally, we hypothesize that gen-der and ethnicity moderate the effects ofparental and friends’ influences, and thereforemultiple group comparisons of the structuralmodels for the different gender and ethnicgroups will be conducted to examine genderand ethnic differmcw in w&l inflwnr~c

,,i *

i

252 JOURIGAL OF HEALTH AND SOCIAL BEHAVIOR

FIGURE 1. Structural Model of Smoking Influences

. .METHODS.‘..

-.-+

Subjects

The data for this study were collected aspart of the Television, School, and FamilyProject (TVSFP, Flay et al. 1988) smokingprevention program in Southern California. ITime 1 (Tl) data were collected in 1986 from6,695 seventh-grade students in 169 class-rooms in Los Angeles Country (representing35 public schools in 4 school districts) and 67classrooms in San Diego County (represent-ing 12 public schools in 2 school districts). Ofall the subjects, 49.6 percent were females,15.5 percent were,Blacks, 32.5 percent wereWhites, 35.5 percent were Hispanics, and16.5 percent belonged to other ethnic groups(half were Asians, referred to as “Others/Asians” in the rest of the text). Time 2 (T2)data were collected 15 months later than Tl,when 4,896 (73.13%) of Tl subjects wererecontacted as eighth-graders.

For the purpose of this study, only thosewho reported “never smoking” and smokingonly one or less than one cigarette at Tl were-included in the analyses. We excluded thosestudents who reponed a disruptive familystructure (e.g., single parent or no parent)from the analysis.’ The eligible sample for theinitiation analysis consisted of those students

. .

who reported never smoking at Tl andprovided information on all of the items ofinterest. The final sample for the initiationanalysis had 1,402 subjects (55.07% werefemales: 12.4% were Blacks, 38.0% wereWhites, 30.0% were Hispanics, and 22.2%were Others/Asians).

The eligible sample for the escalationanalysis consisted of those students whorepotted smoking one or less than onecigarette at ?;,I (experimenters) and providedvalid information on all of the items ofinterest. The final sample for the modelestimate was 527 (54.08% were females,11.05% were Blacks, 37.14% were Whites,30.48% were Hispanics, and 21.33% wereOthers/Asians).

ideasuremetu of Variables

Exogenous Variables. The exogenous vari-ables, friends’ smoking and parental smok-ing, were measured at Tl. Friends’ smokinghad two indicators: friends’ trying smokingand friends’ regular smoking. Friends’ tryingsmoking was assessed by asking the studentshow many of their ten closest friends hadtried a cigarette, and friends’ regular smokingwas measured by asking how many of theirten closest friends IKIKIIIV ~rnnkd 2t lfncr nnp

SMOKING INFLUENCE 253

cigarette a week (from’1 = none to 6 = 8-10friends). Because the responses to the twoitems were highly skewed, they were recodedas dichotomized variables, with 0 = none offriends try smoking (smoke at least onecigarette a week) and 1 = one or morefriends try smoking (smoke at least onecigarette a week). Parental smoking also hadtwo indicators: fathers’ smoking and mothers’smoking. Fathers’ smoking was measured byasking the students how often their father (orstepfather) smokedcigarettes (where 1 = I donot have a father or stepfather, 2 = hedoesn’t smoke, to 7 = he smokes 2 or morepacks a day). The same measure wasdeveloped for mothers’ smoking. Again,these variables were recoded as 0 = doesn’tsmoke currently and 1 = smokes currently.

they thought they would ever smoke in thefuture and if they thought they would ever askanother person to let them try a Cigarette(responses to both items ranged from 1 =definitely would to 6 = definitely wouldnot). These two items served as two indica-tors for the construct of smoking intentions.

general will contract lung career or heart .

Endogenous Variables. Endogenous vari-ables included negative outcome expecta-tions, perceived friends’ approval of smok-ing, perceived parental approval -of smoking,intentions to smoke, initiation of smoking (inthe model for smoking initiation), andescalation of smoking (in the model forsmoking escalation). All the endogenousvariables were measured at Tl except initia-

tion and escalation of smoking. Negativeoutcome expectations were measured byasking four questions concerning the healthconsequences of cigarette smoking: howlikely it is (from 1 = more than 80% likely to5 = less than 20% likely) that smokers in

Initiation of smoking was measured bycontrasting self-reported lifetime smoking atTl and T2. The students were asked: Howmany cigarettes have you smoked in yourwhole life? (where 1 = none at all, 2 = Ihave had only one puff, 3 = part or all of onecigarette, 4 = 2 to 4 cigarettes, 5 = 5 to20cigarettes, 6 = 1 to 5 packs, and 7 = morethan 5 packs). Initiation of smoking wascoded as 0 (remained never smoking) and 1(transition from never smoking to initiation ofsmoking). That is, if participants werenever-smokers at Tl but became smokers byT2, they were classified as “initiators.” Ofthe 1,402 never-smokers at Tl, 450 (32.1%)became “initiators” by T2.

the, transition from experimental smokine

Escalation-of smoking was also measuredby contrasting self-reported lifetime smokingat Tl and T2. Escalation of smoking wascoded as 0 (remained experimenting withsmoking) and 1 (transition to a higher level ofsmoking). That is, if participants smokedonly one puff or part or all of one cigarette atTl (experimenters), but smoked two to fourcigarettes or more by T2, they were classifiedas “escalators.” This study only considered

disease; how likely it is- that smokers in.general might die from lung cancer or heartdisease; how likely it is that you wouldcontract lung cancer and heart disease if yousmoked; and how likely it is that you woulddie from those diseases if you smoked. Thesefour items served as four indicators for theconstruct of negative outcome expectations.

(smoking all of one cigarette or less) at Tl to’a higher level’of smoking (smoking more thanone cigarette) at T2. We did not considerother kinds of smoking transitions in thispaper. Of the 527 experimenters at Tl, 265(50.3%) became “escalators” by T2.

Analytic SrrategyPerceived friends’ approval of smokingwas assessed by asking the students howmany of their ten closest friends wouldapprove if they smoked cigarettes (from 1 =none to 6,U= 8-10 friends). Perceived parentalapproval of smoking was measured by askinghow students’ parents would feel about theirsmoking (from 1 = definitely disapprove to 5= definitely approve). Refusal self-efficacywas assessed by asking the students how hardit would be to refuse a friend’s offer ofcigarettes (from 1 = extremely hard to 6 =not at all hard). Intentions were measuredwith twn item< The cttvbntr WP~- -CL-A :C

Structural equation modeling was the majoranalytic strategy in this study (Joreskog and!&born 1988). Because most of the measureswere ordinal, we used polychoric or polyser-ial correlations. We then used the method ofweighted least squares (WLS) to estimate theparameters for the total sample and fordifferent genders. Multiple group compari-sons were conducted to examine the differ-ence in parameter estimates for the structuralequation models of smoking initiation among,I:&-,-,... . . . . t 1 1. ^.

., ._...._..: .,,.,. ; . . . . . . ,. I_,. .;-: :., ;‘. : .,

254 JOCRh’AL OF HEALTH AND SOCIAL BEHAVIOR

‘some of the ethnic groups (e.g., Blacks andOthers/Asians) consisted of relatively smallsamples, an asymptotic variance and covari-ante matrix could not be constructed. As aresult, WLS could not be used in the modelestimates for these groups. Consequently, weused a mtiimum likelihood solution (MLS)in the structural model estimates for all theethnic groups and for the multiple groupcomparisons. Structural model comparisonscould not be conducted for the models ofsmoking escalation because of the limitedsamples in the subgroups.

RESULTS

Smoking Initiation Mean Analysis

.!:,.

?.-.:.::

Table 1 shows the group means for thegender and ethnic groups on the 14 variablesused -in constructing the structural modelspredicting smoking initiation. Overall, therewere no significant differences in the meanlevels of most variables between males andfemales except that males reported slightlyhigher levels of friends’ and parental approvalof smoking than females (p < .05).

A one-way analysis of variance for the fourethnic groups on each of 14 variables wasconducted to test the difference in groupmeans. The reslilts indicated that significantlymore friends had tried cigarette smokingamong Whites than among Others/Asians butthe number of regularly smoking friendsappeared to be similar across the four groups.A significantly higher percentage of Blacksreported father smoking compared to Whites(41% vs. 28%) and mother smoking com-pared to Whites, Hispanics, and Others/Asians (37% vs. 20%. 17% and !4%+,respectively). With regard to the negativeoutcome expectation variables, Whites scoredsignificantly lower on two of the items thanother ethnic groups. This suggested that

’ Whites had fewer negative beliefs about thehealth consequences of cigarette smokingthan other groups. There were no significantdifferentes in the perceived numbers offriends who would endorse adolescents’ ownsmoking. but Hispanics reported higher levelsof parental approval of smoking than otherethnic groups. In. addition, both Blacks andWhites scored significantly higher on refusalself-efficacy than Hispanics and Others/Asians.

On the smoking intention variables. Whites

reported significantly lower intentions . tosmoke in the future than Blacks and Hispan-ics, and Others/Asians reponed significantlylower intentions to ask another person to letthem try a cigarette than Hispanics. Finally,there were no -significant differences in thesmoking initiation rates across the four groupsusing pairwise comparison (31% for Blacksand Whites, 37% for Hispanics, and 27% forOthers/Asians).

Overall Smoking InitiationStructural Modeling

A total of 1,402 subjects were used in tsti-mating the structural model predicting smokinginitiation. We used weighted least-square meth-ods to estimate structural coefficients and modelfit. We estimated an initial model (full model)with all the hypothesized pathways depicted inFigure 1. Then a final model was estimatedwith only statistically significant paths re-tained. The initial model rovided a good rep-resentation of the data (xP(57) = 62.07, p =.300, Goodness-of-Fit Index (GFI) = ,997).After eliminating nonsignificant paths, the finalmodel (~‘(65) = 70.91, p = .287, GFI =.996) fitted the data well.

Figure 2 shows parameter estimates for thefinal structural model of smoking initiation. Theresults generally supported our hypotheses. Themodel showed that friends’ smoking had bothdirect and indirect effects on smoking initia-tion. The indirect effects were channeled throughsmoking intentions and negative cutcome ex-pectations. As expected, exposure to smoking-friends directly triggered initiation of cigarettesmoking. Also, exposure to smoking friendselevated adolescents’ smoking intentions, whilealso diminishing adolescents’ negative out-come expectations about cigarette smokingwhich, in turn, elevated adolescents’ smokingintentions. Although friends’ smoking had astrong effect on friends’ approval of smoking,this effect did not seem to carry through tointentions. Parental smoking had only indirecteffects on smoking initiation. These effects weremediated through smoking intentions, negativeoutcome expectations, and parental approval ofsmoking.

Overall Smoking EscalationStructural Modeling

A total of 518 subjects were eligible formodel estimation. Again. we used weighted

TABLE 1. Means and Standard Devialions of Measured Variables for the Different Gender and Ethnic Groups

,Gender Groups Ethnic Groups

r-

Males Rmalcs5

Black White tlispanic Other/AsianI-test P- value

i$m

(N = 629 ) (N = 771 ) (N = 174); (N = 533 ) (N = 378 ) ( N = 311)

E.wgenuus VariablesI. kiends’ Smoking

Pricnds’ trying smoking (O-l)Priends’ regular smoking (O-l)

!. Parental SmokingPather smoking (O-l)Mother smoking (O-l)

.43 (.50)

.20 (.40)

.3l (.46)

.20 (.40)

.47 (SO)

.2l (.41)I .54.79

.40 (.49)

.2l (.4l)

.4l (49).37 (.49)nh

.59 (SO)” .47 (.50)

.I8 (.38) .25 (.43)

.28 (.45)’

.20 (40)’.34 (.47).I7 (.38)a

38 (.49)’.I9 (.40)

4.97*+.2.26

.36 (.48) 4.52.’

.I4 (.35)= 13.94**+

\.

.34 (.47)

.20 (.40)1.10.09

Cadogenorts VariablesNegative Outcome Expectationsa. Smokers contract lung cancer/heart discasc (l-5) 2.49 (I .30)b. Smokers die from lung cancer/heart disease (l-5) 2.00 (1.24)c. ‘You contract lung canccr/hcart disease

if you smoke.’ (l-5)d. ‘You die from heart disease/lung cancer

2.48 (I .28)

if you smoke.’ (l-5)Perceived Friends’ Approval of Smoking (i-6)Perceived Parental Approval of Smoking (l-5)Refusal Self-Efficacy (l-6)Smoking Intentionsa. Smoke in the future (l-6)h. Ever ask another person lo let them try a

2.31 (1..11)1.64 (1.31)I .40 (I .06)3.89 (I .54)

1.67 (1.00)

2.63 (1.16) 2.14*2.12 (1.24) I .77

2.72 (I .39)2.29 (I .35)’

2.68 (1.41)

2.61 (I .26)’1.48 (1.19)1.26 (.73)a4.14 (1.48)”

2.47 (I .20) 2.55 (1.21) 2.67 (1.18) 2.81’1.88 (1.28)‘” 2.15 (l.27)h 2.14 (1.21)’ 6.61+*’

2.43 (1.19) 2.50 (I .27) 2.49 (I .26) 1.74 *2.50 (I .24) .I55

2.34 (I .25) 51I.50 (1.11) 2.07*1.29 (.90) 2.28*3.80 (1.54) I.13

1.58 (.9l) I .82 1.87 (1.13)”

2 .46 (1.31)b1.65 (1.29)I .53 (I .27)&3.70 (I .63)”

2.13 (l.23)*b1.59 (1.20)I .24 ( .78)b3.99 (I .42)‘d

I .47 (.84)“b

2.41 (1.26) 8.40*+*1.49 (1.12) 1.30I.31 (.9S)’ 7.38**+3.60 (I .62)M 7.60***

1.65 (1.08) 9.79**+I .70 (.88)h

cigarette (i-6) 1.54 (.94) 1.47 (.83) 1.44 1.57 (1.03) 1.47 (.85) 3.69*

‘\

I .6l (.93)’ 1.41 (.78)”

* D < .05: ** n < .Ol: *+* n < .cOt.A$ N;rcs: Two subjects did not provide-information on gender. and six subjects did not provide information on ethnicity. The total sample sizes for gender and ethnic analyses were I.400

d 1,396, respectively. Same letters in a row indicate that these ethnic groups have significantly different means.

.

:_._._. _^L_. .._ ..&, _._-.___.__.____ ,___-___ . .._.. ~I.... ..; ~... . . . :. . . :--. :. . . .._. .: :, .:: . . . . :: .: . . . . . . .._. : ._:., _ :

256 JOURKAL OF HEALTH AND SOCIAL BEHAVIOR

FIGURE 2. Smoking Initiation Structural Model for the Total Sample (Significant values forcoefficients are: ‘p < .05; ** p < .Ol; *** p < .OOl).

least-square methods to estimate structuralcoefficients and model fit. The initial model(full model) with all hypothesized pathwaysfitted the data well (x-(59) = 46.39, p =.883, G F I = .993). After deleting thenonsignificant paths, the final model wasobtained (~‘(68) = 53.98, p = .892, GFI =.991).

Figure 3 shows the parameter estimatesfrom the final structural model predictingsmoking escalation. Neither friends’ smokingnor parental smokin,0 had direct effects onescalation. The indirect effects of friends’smoking were channeled through negativeoutcome expectations, perceived friends’ ap-proval of smoking, and refusal self-efficacy.As anticipated, exposure to smoking friendsdiminished adolescents’ negative outcomeexpectations, fostered their perceived friends’approval of cigarette smoking, and erodedtheir refusal self-efficacy, all of whichindirectly affected adolescents’ escalation ofsmoking through smoking intentions.

Parental smoking also had significanteffects on negative outcome expectationsabout cigarette smoking. Additionally, paren-tal smoking improved adolescents’ perceivedapproval of smoking from their parents, buthad no significant effects on refusal self-efficacv.

In order to compare the relative influencesof friends and parents at different stages ofsmoking, we presented decompositions of theeffects of friends’ and parental smoking onadolescents’ smoking initiation and escalationfrom the estimation of the full models (Table2). Borh friends and parents had stronger totaleffects on smoking initiation than on escala-tion. In addition, the relative impact offriends and I,arents seemed to vary with thestages of smokin,.0 It was obvious that friendshad stronger direct and total effects oninitiation than parental smoking, but thepicture was less clear for escalation. Bothfriends and parents had very weak effects onescalation and the influences were onlyindirect.

Gender and Erhrlic Differences ill Friends ’and Parenral Injluences

Multiple group comparisons were con-ducted to test for gender and ethnic differ-ences in the smoking initiation structuralmodels. To compare the fit of the models fordifferent gender and ethnic groups, weevaluated four models with different equiva-lence constraints for the samples, separatelyfor gender and ethic comparisons (Table 3).

. .

.., _ ..; 1.___. _ .--_ -...--. _ __. . . _ _ .i -. . ..Y. . :. - . . ..__.. .; .:. _. .. -. : .:..: _.

257 ..: . . .: . . ..-___.

SMOKING INFLUENCE I

~GURE 3. Swing &caintion Structural Model for the Total Sample (Significant values forco&icients are: +p < .05; l *p < .Ol; ***p C .oOl).

II!!42;- ,

,

The ftrst model applied the basic model to thesamples with equivalence constraints only onthe measurement model. The second modelincluded equivalence constraints on the mea-surement mode1 and all the paths fromfriends’ smoking to the endogenous latentconstructs. The comparison of Models 1 and2 tested if there were significant differences inall the paths from friends’ smoking toendogenous variables across the four groups.The third model included equivalence con-

TABLE 2. Decomposition of the Effects ofFriends’ Smoking and ParentalSmoking on Adolescerits’ Initiationand Escalation of Smoking

Initation EscalationO f of

Smokine Smokine

The Effects of Friends’ SmokingDirect Eff+xts .280*** -Indirect Effects .086*** .o#***Total Effects .366*** .064***

The Effects of Parental SmokingDirect EffectsIndirect Effects .087*** .041*Total Effects .087*+' .041*

* p < .os; ** p c .Ol; *** p < ml.‘-’ The direct paths were eliminated in the final

models because they were not significant.

:.

straints on the measurement model and thepaths from both friends’ smoking and parentalsmoking to all endogenous variables. Thecomparison of Models 3 and 2 tested thedifferences in all the paths from parentalsmoking across the groups. The last modelwas estimated with equivalence constraints onthe measurement model and all structuralparameters. The comparison of Models 4 and t3 tested the differences in the mediating paths(including the paths from negative outcomeexpectations, perceived friends’ approval ofsmoking, perceived parental approval ofsmoking, and refusal self-efficacy to smokingintentions, and the path from smokingintentions to initiation).

Gender Comparisons: Goodness-of-fit testswere conducted by comparing the relative fitof the data provided by nested models (Table3 shows the chi-squares associated with thefour models). We compared Models 2 and 1and did not find a significant increase in thec h i - s q u a r e (x*(5) = 5 . 4 1 , p > .05),suggesting that there were no significantgender differences in the paths from friends’smoking to endogenous variables. The com-parison of Models 3 and 2 also yielded anonsignificant result (x2(5) = 4.48, p >

.. :..~-----.. . ____ .: I.--..

258 JOURIKAL OF HEALTH AND SOCIAL BEHAVIOR

TABLE 3. Multiple Group Comparibn of Structural Models of Significant Others’ Influences on . ‘.Adolescents’ initiation of Smoking for the Different Gender and Ethnic Groups

4; ;.:2:

Gender Groups Ethnic Groups

Chi-square incremental $ Chi-square Incremental xaModels (d.f.1 (d.f.) (d.f.) (d.f.)

1. Equivalent Constraints on Factor Loadings . 107.8 1 (I 30) 321.24 (266)2. Equivalent Constraints on Factor Loadings

and all Paths from Friends’ Smoking to theEndogenousVariables 113.22 (135) 5.41 (5) 347.83 (28 I ) 26.59 (15)*

3. Equivalent Constraints on Factor Loadingsand all Paths from both Friends’ Smokingand Parental Smoking to all EndogenousVariables 117.70 (140) 4.48 (5) 362.95 (296) IS.12 (15)

4. Equivalent Constraints on Factor Loadingsand all Structural Parameters 140.53 (145) 22.83 (5)*** 378.77 (311) 15.82 (15)

l p < .os; ** p < .Ol; *** p < .oOl.

::.:,,

.05), indicating that males and females didnot differ on the paths from parental smokingto all the endogenous variables. Finally,comparison of Models 4 and 3 showed thatthe mediating paths between friends’/parentalsmoking and initiation of smoking weresignificantly different for males and females(x’(5) = 22.83, p < .001).3

Since we found significant differences in themediating pathways for males and females, fur-ther analyses were conducted to examine whichpath(s) accounted for the differences. Using mul-tiple group comparisons as described above,we conducted the analyses path by path. A totalof five mediating paths (outcome expectationsto intentions, friends’ approval to intentions,parental approval to intentions, refusal self-efficacy to intentions, and intentions to initia-tion) were compared separately, for males andfemales. The results showed that the differ-ences lay in the path from parental approval ofsmoking to smoking intentions, which was sig-nificant for females, but not for males, indicat-ing that parental approval mediated the effectsof parental smoking only in females, not inmales.

Erhnic Comparisons: The chi-squares asso-ciated with the four nested models for theethnic groups are also shown in Table 4. Wecompared Models 2 and 1 and found asignificant increase in the chi-square (x2(15)= 26.59, p = .03), suggesting significantdifferences in the paths from friends’ smokingto the endogenous variables across the fourgroups. The comparison of Models 3 and 2yielded a nonsignificant result (x2(15) =15.12, p > .05), indicating that the pathsfrom parental smoking to the endogenousvariables were not significantly different

among the four ethnic groups. Finally,comparison of Models 4 and 3 showed. thatthe mediating paths between friends’ andparental smoking and initiation of smokingwere also not significantly different amongthe four groups (x2(15) = 15.82, p > .05).

Since we found significant differences infriends’ pathways among the four groups,further analyses were conducted to examinewhich path(s) accounted for the differences.Using the multiple group comparisons asdescribed above, we conducted the analysespath by path. A total of five paths (friends’smoking to initiation, friends’ smoking tointentions, friends’ smoking to outcomeexpectations, friends’ smoking to friends’approval of smoking, and friends’ smoking torefusal self-efficacy) were compared sepa-rately. The results showed that the differences ;.

Alay in two paths: (1) the path from friends’smoking to negative outcome expectations,which was significant for Hispanics but notfor other groups, and (2) the path fromfriends’ smoking to refusal self-efficacy,which was significant only for Whites.

Table 4 presents the decompositions ofparental and friends’ effects for the differentgender and ethnic groups. The total effects offriends’ smoking on initiation were similar formales and females, but males had more directeffects and females had more indirect effects.Parental smoking had only indirect effects oninitiation, but the effects appeared to begreater for females than for males.

In regard to the four ethnic groups, it wasclear that friends’ smoking had much strongereffects than parental smoking across all thegroups. Friends’ smoking had significantdirect effects on initiation for all groups. but

/

,I i,

.: ;

..,...

.

.-

: :,..-: . . . . . Il.-l_...:-.ll‘...crr _-._ ;.;;l.--,-~,~;::~Y.i.l;.;i:.c . i ._...-. 1 I . . :-: ,:

SMOKING INFLUENCE 259TABLE 4. Decomposition of the Effects of Friends’ Smoking and Parental Smoking on

Adolescents’ Initiation for Different Gender and Ethnic Groups

Gender Groups Ethnic Groups

M&S Females Black White Hispanic Other/Asian(N = 629) (N = 731) (N = 174) (N = 533) (N = 378) (N = 311)

77ae Effccrs ofFriends ’ Smoking

Direc! Effects .316*** .235*+’ .366* .192* .343=** .325***Indirect Effects .042** .135*** .128*** .091**Total Effects .358*** .370’“” .366* .320*** .434***. .325***

The Effects ofParental Smoking

Direct Effects - -Indirect Effects .075* .127*** .Oll .153*Total Effects .075* .127*** .Oll :153* - *

*p<.05;**p<.01;***p.<oo1.‘-’ These paths were eliminated in the final models because they were not significant.

the indirect effects were significant only forWhites and Hispanics. Parental smoking hadsignificant indirect effects on initiation onlyfor Hispanics.

DISCUSSION AND CONCLUSIONS

The purposes of this study were to examinethe differential influence of parental smokingand friends’ smoking on adolescents’ initia-tion and escalation of smoking, and to exploregender and ethnic differences in parental andfriends’ influences on adolescents’ smokingbehavior. We began this paper by raisingseveral different issues about the effects ofothers’ behavior on adolescent smoking. We

employed structural equation models to ad-dress these issues.

The first issue addressed concerns aboutwhether others’ behaviors affected adolescentsmoking by direct modeling or indirectlythrough mediating variables. Our resultsclearly show that friends’ smoking has bothdirect and indirect influences on adolescentinitiation of smoking, but only indirect effectson escalation. Parental smoking has onlyindirect effects on initiation and escalation,channeled through mediating variables.

The second issue concerns the relativeinfluence of parental and friends’ smoking onadolescents’ smoking behavior. Our findingssuggest that friends’ smoking has strongerinfluences on smoking initiation than doesparental smoking, and the pattern becomesmore pronounced when ethnic groups areexamined separately.

The third issue concerns the different

effects of others’ behavior on different stagesof smoking behavior, that is, initiation andescalation of smoking. Our study shows thatfriends’ smoking has stronger influences oninitiation than on escalation. In addition,friends’ smoking can directly influence initia-tion of smoking, but its influence on smokingescalation is only indirect. Moreover, itsindirect effects on initiation are channeledthrough negative outcome expectations andintentions, while the indirect effects offriends’ smoking on escalation are channeledthrough perceived friends’ approval of smok-ing and refusal self-efficacy, in addition tonegative outcome expectations and intentions.Parental smoking has only indirect effects onsmoking initiation and escalation.

Finally, the study addresses gender andethnic differences with respect to the effect offriends’ and parental smoking on adolescentinitiation of smoking. Multiple group compar-isons of the structural model predictingsmoking initiation between males and femalesshow that parental approval of smokingmediates the effects of parental smoking forfemales, but not for males. It is also shownthat the effects of parental smoking arestronger for females than for males. Theseresults are consistent with previous findingsthat girls might be more susceptible to socialinfluences, especially parental influences,than boys (Ensminger et al. 1982).

Multiple group comparisons amongWhites, Blacks, Hispanics, and Others/Asians revealed that there are some signifi-cant differences in friends’ influences. Inparticular, aside from the common significantpathways such as the paths from friends’

:+;:i..::.’ :..

: ,:.:

..,

.,_ ..____- _ -_.. ‘ .__.____ _________._.._ . . ..__...-_.... L.. ~._. , . . . . __, . .._

260 JOL’RNAL OF HEALTH AND SOCIAL BEHAVIOR

smoking to initiation and friends’ approval ofsmoking, other unique pathways are specifi-cally relevant to the prediction of smokinginitiation in different ethnic groups. ForWhites, friends’ smok.ing has significanteffects on adolescents’ refusal self-efficacy;for Hispanics, it has significant effects onnegative outcome expectations. In addition,friends’ smoking has both significant directand indirect effects on initiation for Whitesand Hispanics and only significant directeffects for Blacks and Others/Asians.

The present research is not intended to testspecific theories. Instead, we construct hy-potheses from different perspectives andvarious theories of social influences, whichwe then integrate into a unified model toclarify the process of the influences offriends’ smoking and parental smoking onadolescents’ smoking behavior. Our resultsprovide support for most of the theoreticalperspectives integrated in the overall struc-tural model and raise some questions aboutothers.

Consistent with social learning/cognitiveperspectives, our models reflect two basiclearning processes; that is, modeling andcognitive mediation. Modeling is a learningprocess that occurs by observing others anddirectly copying their behaviors. This processproduces direct effects in our model. Con-versely, cognitive mediation is a learningprocess through which others’ behaviorsaffect the antecedent determinants of behav-iors first. such as outcome expectancies andself-efficacy, which, in turn, affect adoles-cents’ own behaviors. Our study underscoresthe importance of the roles played by bothmodeling and cognitive mediation in adoles-cents’ smoking behavior. However, the twoprocesses may occur in a different manner atdifferent stages of behavior. Specifically,modeling friends’ smoking behavior appearsto dominate the process of adolescents’smoking initiation, whereas the influence offriends’ smoking on escalation mainly resultsfrom the mediating effects. These findingsspeak directly to the concept of sociallearning postulated by Krohn et al. (1985),that initial engagement in smoking behavior islikely to be a result of imitation of rolemodels. After initial use, social and nonsocialreinforcement is more likely to be experi-enced, and therefore, maintenance of thebehavior is less dependent on imitation ofnthers’ hehnvior. F u r t h e r m o r e . our studs

points out that adolescents’ learning processesmay vary with the sources of the influences.In particular, adolescents are more likely tocopy their friends’ smoking behavior directly,while parental behaviors tend to exert theirinfluences by affecting adolescents’ smokingattitudes and beliefs.

The present study also provides consider:able support for the theory of reasoned action(Fishbein and Ajzen 1975; .Ajzen and Fish-bein 1980) and the theory of planned behavior(Ajzen 1985). Specifically, our models dem-onstrate a strong linkage between intentionsand actual smoking behavior and the predic-tive role of outcome expectations, perceiLedfriends’ and parental approval of smoking,and refusal self-efficacy. It should be noted,however, that operationalitation of variablesin our study does not fully correspond toAjzen’s theory. For instance, the concept ofsocial normative beliefs, an important con-struct in the theory of planned behavior, is notincluded in our models. Instead, we use theconstructs of perceived friends’ and parentalapproval, which are only partial measures ofsocial normative beliefs, as mediating vari-ables. Similarly, regarding another importantconstruct, attitudes toward behavior, weinclude only the health consequences compo-nent of smoking attitudes. The psychosocialpart of smoking attitudes may be moreimportant in predicting adolescents’ smokingbehavior. In addition, our models includeonly refusal self-efficacy, even though theconcept of use self-efficacy, which representsadolescents’ belief;. in their ability to obtainand use cigarettes, may be equally important.Moreover, Ajzen and Fishbein (1980) haveemphasized the importance of correspondencebetween the measures of intention andbehavior and recommended short time periodsfrom intention to behavior. In this particularstudy, our outcomes (smoking initiation andescalation) do not have a definite time frame,but could have occured at any time during the15month follow-up. Nevertheless, the stronglinkage between intentions and behaviordemonstrated in this study can be viewed asfurther evidence for the importance of themediating role of intentions postulated byAjzen and Fishbein.

Finally, our models suggest that there existsignificant paths from parental and friends’smoking to adolescent intentions to smoke.These findings are counter to Ajzen andFishhein’s hnsic nntion that intentions are rhe

-: -2 .-_, A_._ ‘__~ ___._ _. ._....... I,.-; _.._. _... :... . . . . . . -. . . . . . .:: ... . .

i!-_’ SMOKING INFLUENCE 261

exclusive determinant of behavior and that Another constraint of this study arises fromintentions. in turn, are only determined by the fact that we exclude those students whoattitudes, social normative beliefs, and self- came from disrupted families. Consequently,efficacy. The same challenge has been raised this research may be generalizable only to

by other researchers. For instance, Bentler those students who have an intact family.and Speckart (1979) found that prior experi- Previous studies have documented that familyence with drugs could directly affect current structure is an important predictor of drug usedrug use. (e.g., Baumrind 1985; Kandel et al. 1978). In

Generally speaking, this research is consis- particular, these studies found that children:. tent with the socialization theory. As dis- and young adolescents from homes with_’

cussed by Kandel, Kessler, and Margulies single or divorced parents were at risk for(1978:73), the use of drugs by adolescents subsequent marijuana use. In this regard, the“provides a unique opportunity for studying social influences of cigarette smoking forprocesses of socialization, whereby individu- children from an intact family might beals learn values and behaviors.” Our results different from those from a disrupted family..are consistent with previous research on Further research is needed to test this/adolescent socialization into marijuana, alco- hypothesis.

,: hol, and hard drug use (Kandel et al. 1978; Finally, it should be pointed out thatKandel and Adler 1982; Kandel and Andrews siblings, especially older siblings, can be1987) on the following points: (1) friends can considered as an important source of influ-influence adolescents’ behavior by direct ence on adolescent substance use. Siblingsmodeling, cognitive-mediation, or both; (2) in have been characterized as a special categorygeneral, friends have more influence than of peer influence and they can act as roleparents in early adolescence; and (3) the models by reinforcing certain behavioral

! influence of parents and peers occurs in a patterns (Needle et al. 1986). Adolescents ’I different manner at different stages of behav- whose older siblings smoke are more likely to

j ior. Furthermore, this study demonstrates that adopt the behavior than those whose siblings_- i the social influences on adolescent initiationI do not smoke (Alexander et al. 1983; de Vries

,-.I

of smoking may differ by gender and ethnic et al. 1990; Goddard 1990; Murray et al._i group, suggesting that socialization processes 1983). Penning and Barnes (1982) have

.:;;may vary by gender and may also vary in reported that older siblings’ influence ondifferent cultures. adolescent substance use is greater than.,;.: .IT. One limitation of this study stems from parents’ but less than peers’ influence.the fact that all measures of smoking were Unfortunately, limited data on siblings doesprovided by adolescents, including measures not permit us to examine this issue in ourof smoking among parents and friends. In models. Further research is needed to explore ;iI’!-1 particular, this study assesses the relation- the socialization role of siblings, relative to -ship between (1) adolescents’ reports of their friends and parents, in adolescent smoking

.-I own smoking, and (2) their perceptions of onset.../ smoking among their parents and friends. In conclusion, the present study points out

This study does not consider whether parents the differential influences of parental smokingand friends actually smoke at the levels and friends’ smoking on adolescent initiationreported by subjects. When addressing social and escalation of smoking. The study under-influences on smoking, the distinction be- scores the importance of differential theoreti-tween perceptions of smoking among others cal perspectives, all of which have important..;,;s. . : and actual smoking among others is critical implications for tobacco prevention andbecause some studies (e.g., Iannotti and intervention (Flay and Petraitis, forthcoming).Bush 1992) have found that the perception From cognitive-affective perspectives, theof smoking among friends is more closely key to preventing tobacco use is throughrelated to an adolescent’s own smoking than persuasive messages which directly targetis the actual number of friends who smoke. tobacco-specific beliefs. The messages shouldConsequently, our reliance on adolescents’ increase adolescents’ expectations regardingperceptions of smoking among others might the adverse consequences of using tobacco._.._; have led us to overestimate the effects of and decrease their expectations regarding theparental and friends’ actual smoking behav- potential benefits of using tobacco. Mean-ior. while, messages should challenge adoles-

-,.:-.- .~. ._ * .-

. ..- ,.,_. . . . . .:,.l--;m.r . . . . .._ i ..-__ u..-.-..--‘-....:-.:~..~-~~-~.:.-:.;.~~ . . -, ~.- .._. _. :>--....

33’ JticR>XL Ul- HEALTH AND SOCIAL BEHAVIOR

cents’ perceptions concerning the normativenature of tobacco use, perhaps by challengingany inflated estimates of prevalence andperceived approval of Lobacco use amongtheir peers. From social learning perspectives.a major key to prevention lies in makingtobacco-using role models less salient andtobacco-abstaining role models more salient.In addition, teaching refusal skills andenhancing adolescents’ refusal skills arecrucial as well. Finally, the findings of genderand cultural differences in smoking influencesfrom this study suggest that smoking preven-tion and intervention programs need to besensitive to culture and gender.

NOTES

1. Although the study intervention influenced somemediating variables, the intervention did not in-fluence cigarette smoking or other drug use (Flayet al. forthcoming). Thus. smoking rates in theintervention group did noi differ significant&from that of the control’group at baseline andafter intervention. Consequently, the present re-search collapsed the samples from the interven-tion and the control groups. In addition, sinceall of the mediating variables used in the presentstudy were measured at baseline, the interven-tion was not considered to have any significanteffects on the model estimates. To confirm this,we applied the same structural models used inthe study to the intervention and control groupsseparately and employed multiple group com-parisons to test if there was any significant dif-ference in the structural models between the in-tervention and control groups. The results showedthat, for the smoking initiation structural model,the relative fit of the model with equivalenceconstraints on the measurement model was xa= 207.09 (d.f. = 151) and the relative fit of anested model with equivalence constraints onboth measurement and structural models was x2= 231.09 (d.f. = 166). The difference in the fitof the data provided by the two models was notsignificant (x’ = 24.00 (d.f. = 15). p > .05),indicating that there were no significant differ-ences in the smoking initiation structural modelsbetween the intervention and control groups. Inregard,to the smoking escalation structural mod-els, the relative firs of the two nested modelswere x2 = 155.84 (d.f. = 151) and x2 =164.69 (d.f. = 166). respectively. The differ-ence in the relative fit between the two modelswas not significant either (x’ = 8.85 (d.f. =15). p > .05).

2. Based on the responses to the questions aboutparental smoking (discussed below in “measure-ment of variables”). when we allowed for “do

not have a. father or stepfather” and “do nothave a mother or stepmother” responses. weconducted additional analyses to examine ethnicdifferences in family structure. The resultsshowed that among Blacks. 2.25 percent re-ported not having a father or a mother, 7.09per&t did not have a father, and 2.58 percentdid not have a mother: among Whites, .48 per-cent did not have a father or a mother. 5.25percent did not have a father, and .99 percentdid not have a mother: among Hispanics, 1.10percent did not have a fathei or a mother, 7.68percent did not have a father. and 1.46 percentdid not have a mother; among Others/Asians,the corresponding percentages were 1.02 per-cent, 5.22 percent, and 1.59 percent, respec-tively. In order to control for the confoundingeffects of family structure, our analysis only in-cluded those who had an intact family (havingboth parents). Consequently, significantly higherproportions of Blacks and Hispanics (especiallyBlacks) were excluded from this study.

3. One anonymous reviewer suggested that oneadditional comparison of a new model withequivalence constraints on measurement modeland paths from parental influences againstModel 1 was needed to estimate the “pure”parental influences. We took this suggestionand the results remained unchanged. Specifi-cally. the difference of chi-squares between thenew model and Model 1 was 4.19 with fivedegrees of freedom (p > .05) and the differenceof chi-squares between the new model andModel 3 was 5.70 with five degrees of freedom(p > .05). These results were very close tothose shown in Table 3. Meanwhile, weconducted similar analyses for ethnic groupcomparisons. The difference of chi-squaresbetween the new model and Model 1 for thefour ethnic g-ups was 15.88 with 15 degreesof freedom (p > .05) and the difference ofchi-squares between the new model and Model3 was 25.83 with 15 degrees of freedom (p <.05). These results were also similar to those inTable 3.

REFERENCES

Alexander, N.M., R. Callcott, A.J. Dobson, G.R.Hardes. D.M. Lloyd, D.L. O’Connell, and S.R.Leeder. 1983. “Cigarette Smoking and DrugUse in Schoolchildren: IV-Factors Associatedwith Changes in Smoking Behavior.” Inrerno-rho/ .lourtx~l of Epidemiology 12:59-66.

Ajzen, Icek. 1985. “From Intentions to Actions: ATheory of Planned Behavior.” Pp. 1 l-39 inAction-Control: From Cognition to Behavior,edited by J. Kulh and J. Beckmann. Heidelberg:Springer.

and Martin Fishbein. 1980. UnderstandingArtirudes and Predicting Social Behavior.Englewood-Cliffs. N.J.: Prentice-Hall.

5MORING INFLUENCE

Akcrs, Ronald L.. Marvin D. Krohn,. LonnLanza-Kaduce. and Marcia Radosevich. 1979.“Social Leaning and Deviant Behavior: ASpecific Test of a General Theory.” AmericanSociological Review 44~63655.

Ary, Dennis V., Anthony Biglan. Cheri L. Nautel,Wendy Weissman. and Herbert H. Severson.1983. “Longitudinal Prediction of the Onset andChange of Adolescent Smoking.” Pp. 641-47 inProceedings of Fifih World Conference onSmoking and Health. Winnipeg, Manitoba.

and Anthony Biglan. 1988. “LongitudinalChanges in Adolescent Cigarette Smoking Be-havior: Onset and Cessation.” Journal ofBehavioral Medicine 11:361-82.

Bandura. Albert. 1969. “Social-Learning Theoryof Identificatory Processes.” Pp. 213-62 inHandbook of Sociali:ation Theory and Re-search, edited by D.A. Goslin. Chicago: RandMcNally.-. 1982. “Self-Efficacy Mechanism in Hu-

man Agency.” American Psychologist 37: 122-47.-. 1986. Social Foundations of Though; bnd

Action: A Social Cognitive Theory. New York:Prentice-Hall.

Bauman. Karl E., Lynn A. Fisher, Elizabeth S.Bryan, and Roger L. Chenoweth. 1984. “Ante-cedents, Subjective Expected Utility, and Be-havior: A Panel Study of Adolescent CigaretteSmoking.” Addictive Behaviors 9: 121-36.

Baumrind, Diana. 1985. “Familial Antecedents ofAdolescent Drug Use: A Developmental Per-spective.” Pp. 13-44 in Etiology of Drug Use:Implications for Prevention, edited by C.L.Jones and R.J. Battjes. Rockville, MD: NationalInstitute on Drug Abuse.

Bentler. Peter M. and George Speckart. 1979.“Models of Attitude-Behavior Relations.” Psy-chological Revienv 861452-64.

Biddle. Bruce J., Barbara J. Bank, and MarjorieM. Marlin. 1980. “Parental and Peer Influenceon Adolescents.” Social Forces 58(4): 1057-79.

Castro, Felipe. Ebrahim Maddahian, Michael D.Newcomb,. and Peter M. Bentler. 1987. “AMultivariate Model of the Determinants ofCigarette Smoking among Adolescents.” Jour-nal of Health and Social Behavior 28:273-89.

Charlton. Anne and Valerie Blair. 1989. “Predict-ing the Onset of Smoking in Boys and Girls.”Social Science and Medicine 29(7):813-l 8.

Chassin. :Laurie. Clark C. Presson. Steven J.Sherman. Eric Corty, and Richard W.Olshavsky. 1984. “Predicting the Onset ofCigarette Smoking in Adolescents: A Longitudi-nal Study.” Journal of Applied Social Psychol-ogy 14122443.

Clark C. Presson, Steven J. Sherman.xdl Montello. and John McGrew. 1986.

“Changes in Peer and Parent Influence DuringAdolescence: Longitudinal versus Cross-Sec-

263._

tional . Perspectives on Smoking Initiation.”Developmental Psychology 3:327-34. -

Conrad, Karen M., Brian R. Flay, and David Hill.1992. “Why Children Start Smoking Cigmttes:Predictors of Onset.” British Journal of Ad&c-tion 87:171 l-24.

de Vries, Hein, Margo Dijkstra, Marjolein Grol,Sirnone Seeien. and Gerjo Kok. 1990. "Predic-tors of Smoking Onset and Cessation inAdolescents.” Paper presented. at the SeventhWorld Conference on Tobacco and Health. April1-5. Perth, Australia.

Ensminger, Margret E., Hendricks Brown, andSheppard G. Kellam. 1982. “Sex Differences inAntecedents of Substance Use among Agoles-cents.” Journal of Social Issues 38(2):25-42.

Fishbein, Martin and Icek Ajzen. 1975. Belief,Attitude, Intention and Behavior: An Introduc-tion to Theory and Research. Reading, MA:Addison-Wesley.

Flay, Brian R. and John Petraitis. Forthcoming.“The Theory of Triadic Influence: A NewTheory of Health Behavior with Implications forPreventive Interventions.” In Advances in Med-ical Sociology, Volume IV: A Reconsideration ofModels of Health Behavior Change, edited byG. L. Albrecht. Greenwich, CT: JAI Press.-. Todd Q. Miller. Donald Hedeker, Ohid

Siddiqui. Cynthia F. Britton. Bonnie R. Bran-non, C. Anderson Johnson, William B. Hansen,Steve Sussman, and Clyde Dent. Forthcoming.“The Television School and Family SmokingPrevention and Cessation Project, VIII. StudentOutcomes and Mediating Variables.” PreventiveMedicine.-, Josie R. d’Avemas, J. Allen Best. Mary

W. Kersell, and Katherine B. Ryan. 1983.“Cigarette Smoking: Why Young People Do Itand Ways e-of Preventing It.” Pp. 131-83 inPediatric and Adolescent Behavioral Medicine,edited by P.J. McGrath and P. Firestone. NewYork: Springer-Verlag.-. Bonnie R. Brannon. C. Anderson

Johnson, William B. Hansen, Arthur Ulene,Deborah A. Whitney-Saltiel. Laura R. Gleason.Steve Sussman, Michael D. Gavin, Kimarie M.Glowacz, Debra F. Sobol. and Dana C. Spiegel.1988. “The Television, School, and FamilySmoking Prevention/Cessation Project: I. Theo-retical Basis and Television Program Develop-ment. ” Preventive Medicine 1758547.

Glynn. Thomas J. 1981. “From Family to Peer: AReview of Transitions of Influences amongDrug-using Youth.” Journal of Youth andAdolescence 10:363-83.

Goddard, Eileen. 1990. Why Children StartSmoking: An EnquiT Carried Out by SocialService Division of OPCS on Behalf of theDepartment of Health. London: HMSO.

Huba. George J. and Peter M. Bentler. 1980. “TheFan,? of pppr ?“,A hrl..l. Nl^J-1.. a-..- n.. - ’

.-1.. __ ‘,.’ .:

at Different Stages in Adolescence.” JOWMI ofYouth and Adolescence 9:44965.

Iannotti, Ronald J. and Patricia J. Bush. 1992.“Perceived vs. Actual Friends’ Use of Alcohol,Cigarettes, Marijuana, and Cocaine: Which Hasthe Most Influence?” Journal of Youth andAdolescence 21:375-89.

J‘dreskog, Karl G. and Dag S&born. 1988. Lisrel7.20. Chicago: Scientific Software, Inc.

Kandel, Denise ‘B., Ronald C. Kessler, andRebecca 2. Margulies. 1978. “Antecedents ofAdolescent Initiation into Stages of Drug Use: ADevelopmental Analysis.” Pp. 73-99 in Lo@-tud ina l Research on Drug Use: Empi r ica lFindings and Methodological Issues, edited byD.B. Kandel. Washington, DC: Hemisphere.

and Israel Adler. 1982. “Socialization intoMarijuana Use among French Adolescents: ACross-cultural Comparison with the UnitedStates.” Journal of Health and Social Behovior231295-309.

and Kenneth Andrews. 1987. “Processesof Adolescent Socialization by Parents andPeers.” The International Journal of the Addic-tions 23~295-309.

Krohn, Marvin D.. William F. Skinner, James L.Massey, and Rona!d L. Akers. 1985. “SocialLearning Theory and Adolescent CigaretteSmoking: A Longitudinal Study.” Social Prob-lems 32455-7 1.

Lau. Richard R.. Marilyn Jacobs Quadrel. andKaren A. Hartman. 1990. “Development andChange of Young Adults’ Preventive HealthBeliefs and Behavior: Influence from Parentsand Peers. ” Journal of Health and SocialBehavior 3 1:240-59.

McCaul. Kevin D., Russell Glasgow, H. Kather-ine O’NeiIl. Virginia Freeborn. and Brenda S.Rump. 1982. “Predicting Adolescent Smok-ing.” The Journal of School Health 52~342-46.

Mittelmark, Maurice B.. David M. Murray. RussellV. Luepker. Terry F. Pechacek. Phyllis L. Pirie,and Unto E. Pallonen. 1987. “PredictingExperimentation with Cigarettes: The ChildhoodAntecedents of Smoking Study (CASS).” Amer-ican Journal of Public Health 77:206-208.

Murray. M., A.V. Swan. M.R.D. Johnson, andB.R. Bewley. 1983. “Some Factors Associatedwith increased Risk of Smoking by Children.”British Journal of Psyhiatr-y 24(2):223-32.

Needle. Richard, Hamilton McMubbin, MarcWilsoq, Robert Reineck, Amnon Lazar andHelen !vlederer. 1986. “Interpersonal Influencesin Adolescent Drug Use-The Role of OlderSiblings. Parents and Peers.” The In/crnotiona/Journal of the Addictions 2 1(7):739-66.

. . . ..: . . . . I ;.:.. ‘.. ..: .‘Z. :‘. . .‘. _.. .;. :..I:.]

JOURNAL OF HEALTH AND SOCIAL BEHAVlOR

Newcomb. Michael D.. George 1. Huba and PeterM. Bentler. 1983. “Mothers’ Influence on theDrug Use of Their Children: Confirmatory testsof Direct Modeling and Mediational Theories.”Developmental Psychology 19(5):714-26.

Newcomb. Michael D. and Peter M. Bentler.1983. “Substance Use and Ethnicity: Differen-tial Impact of Peer and Adult Models.” TheJournal of Psychology 120( 1):83-95.

Penning, Margaret and Gordon E. Barnes. 1982.“Adolescent Marijuana Use: A Review.” TheInternational Journal of the Addicrions 17:749-791.

Semmer. Norben K., James H. Dwyer. PeterLippert. Reinhard Fuchs, Paul D. Cleary. andAnita Schindler. 1987a. “Adolescent Smokingfrom a Functional Perspective: The Berlin-Bremen Study.” European Journal of Psyhol-ogy of Education 1 I :387-402.-. Paul D. Cleary. James H. Dwyer,

Reinhard Fuchs, and Peter Lippert. 1987b.“Psychological Predictors of Adolescent Smok-ing in Two German Cities: The Berlin-BremenStudy.” MorbidiF and Mortality Weekly Report36(4s):3-10.

Skinner, William F.. James L. Massey. Marvin D.Krohn. and Ronald M. Lauer. 1985. “SocialInfluences and Constraints on the Initiation andCessation of Adolescent Tobacco Use.” Journalof Behavioral Medicine 8:353-75.

Stacy, Alan W., Brian R. Flay. C. AndersonJohnson. and William B. Hansen. “A Compari-son of Informational, Normative, and individualDifference Factors as Longitudinal Predictors ofAdolescent Smoking.” Department of Preven-tive Medicine, University of Southern Califor-nia. Unpublished manuscript.

Sussman, Steve, Clyde W. Dent. Brian R. Flay.William B. Hansen, and C. Anderson Johnson.1987. “Psyc>iological Predictors of CigaretteSmoking Onset by White. Black. Hispanic, andAsian Adolescents in Southern California.”MorbidiF and Mortality Week!\: Report 36(4s):1 l-6.

Swan. A.V.. R. Creeser. and M. Murray. 1990.“When and Why Children First Start toSmoke.” Irrterrratiorral Journal of Epidemiolog>19:333-30.

Urberg. Kathryn A., Chien-Hung Cheng andShiang-Jeou Shyu. 1991. “Grade Changes inPeer influence on Adolescent Cigarette Smok-ing: A Comparison of Two Measures.” Addic-tive Behaviors l6( l-7):2 l-8.

Waldron. Ingrid, Diane Lye. and AnastasiaBrandon. 199 I. “Gender Differences in Teen-age Smoking.” Women & Health 17(2):65-90.

Brian R. Flay is director of the Prevention Research Center and professor in the School of Public Health.University of Illinois at Chicago. His research interests include smoking cessation. AlDS prevention andviolence prevention.

. . #. .

S~o~il\;c INFLUENCE 265

Fmnk B. Hu is a doctoral candidate in epidemiology in the School of Public Health, University ofIllinois at Chicago. His research interests include etiological processes of drug use and epidemiologicalmethodology.

Ohidul Siddiqui is a research specialist/statistician at the Prevention Research Center, University ofIllinois at Chicago. His research interests include design and analysis of longitudinal studies, andstructural equation modeling.

L. Edward Day is a postdoctoral research fellow at the Prevention Research Center, University ofIllinois at Chicago. His research interests include the prevention of delinquency and adolescent substanceabuse and the effects of community factors on delinquency.

Donald Hedeker is assistant professor of biostatistics. University of Illinois at Chicago. His researchinterests include clustered and longitudinal data analysis of noncontinuous outcome variables.

John Petraitis is associate professor of psychology at the University of Alaska, Anchorage. His researchinterests focus on the etiology of licit and iiiicit substance use by adolescents. .

Jean Richardson is associate professor of preventive medicine at the University of Southern California.Her research interests include the influence of parenting practices on adolescent substance use, academicachievement, and mental health.

Steve Sussman is associate professor of preventive medicine, University of Southern California, and isaffiliated with the Institute for Health Promotion and Disease Prevention Research (IPR). His researchinterests include tobacco, alcohol, and other drug use prediction, prevention, and cessation, particularlyin high risk populations. Also, he has an interest in program development strategies.