expectancies as mediators of the effects of social...

17
Psychology of Addictive Behaviors 1997, Vol. II, No. 1,48-64 Copyright 1997 by the Educational Publishing Foundation 0893-164X/97/S3.00 Expectancies as Mediators of the Effects of Social Influences and Alcohol Knowledge on Adolescent Alcohol Use: A Prospective Analysis Lawrence M. Scheier and Gilbert J. Botvin Cornell University Medical College Expectancies play an important role in the generation of adolescent alcohol use. However, few studies have precisely elucidated their role when specified with other prominent measures of social influences, which may also independently promote alcohol use. Three-year panel data and path-analytic techniques were used to test a model positing that social reinforcement expectancies mediate the effects of per- ceived friends' alcohol use, friends' alcohol attitudes, and knowledge of near-term health effects and alcohol prevalence on both contemporaneous and subsequent alcohol involvement. Evidence of mediation was obtained with both cross-sectional and longitudinal findings. Perceived peer norms had a direct effect on alcohol use, and knowledge of normative alcohol use had a unique long-term protective influence on later alcohol use. Findings are discussed in terms of a 2-pronged prevention model that (a) integrates principles of social learning theory with expectancy-based, cognitive-behavioral change and (b) emphasizes dissemination of age-appropriate alcohol information in programs that aim to reduce alcohol use. A growing body of literature supports the strength of alcohol-related expectancies as important determinants of adolescent drinking behavior (e.g., see Brown, Creamer, & Stetson, 1987; Christiansen & Goldman, 1983; Gold- man, Brown, & Christiansen, 1987; Stacy, Widaman, & Marlatt, 1990). Studies of both youths and adults have demonstrated empiri- cally that perceived outcomes (i.e., anticipated effects) from drinking account for substantial amounts of variation in contemporaneous and Lawrence M. Scheier and Gilbert J. Botvin, Department of Public Health, Institute for Prevention Research, Cornell University Medical College. Preparation of this article was partially supported by a research grant (P50DA-7656) to Gilbert J. Botvin and a FIRST Award (R29-DA08909-01) to Lawrence M. Scheier from the National Institute on Drug Abuse. Portions of these data were presented at the meeting of the Society for Prevention Research, Scottsdale, Arizona, June 1995. Correspondence concerning this article should be addressed to Lawrence M. Scheier, Department of Public Health, Institute for Prevention Research, Cornell University Medical College, 411 East 69th Street, Kips Bay 201, New York, New York 10021. longitudinal drinking patterns (Bauman & Che- noweth, 1984; lessor & lessor, 1977; Mosko- witz, Schaps, Schaeffer, & Malvin, 1984; Sher, Walitzer, Wood, & Brent, 1991; Stacy, New- comb, & Bentler, 1991) and predict well to problem drinking among youth (Chen, Grube, & Madden, 1994; Christiansen, Smith, Roehling, & Goldman, 1989) and young adults (Brown, 1985). Most of these studies have involved a central theoretical premise that, in moderate dosages, alcohol has positive reinforcing proper- ties and that the learned contingency between alcohol and reinforcement or "expectance" is largely responsible for generating drinking behavior (Goldman & Rather, 1993; Lang & Michalec, 1990; Stacy, Leigh, & Weingardt, 1994). In response to this growing body of knowl- edge, prevention programs have developed strategies to deter youth from drinking that focus on changing beliefs regarding the benefi- cial effects of alcohol (e.g., Botvin & Botvin, 1992). A central premise underlying these interventions is that effective barriers to alcohol consumption include restructuring the cognitive linkage between expectancy (i.e., attitude) and behavior (e.g., Stacy, Bentler, & Flay, 1994). 48

Upload: others

Post on 31-Jul-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

Psychology of Addictive Behaviors1997, Vol. I I , No. 1,48-64

Copyright 1997 by the Educational Publishing Foundation0893-164X/97/S3.00

Expectancies as Mediators of the Effects of SocialInfluences and Alcohol Knowledge on Adolescent

Alcohol Use: A Prospective Analysis

Lawrence M. Scheier and Gilbert J. BotvinCornell University Medical College

Expectancies play an important role in the generation of adolescent alcohol use.However, few studies have precisely elucidated their role when specified with otherprominent measures of social influences, which may also independently promotealcohol use. Three-year panel data and path-analytic techniques were used to test amodel positing that social reinforcement expectancies mediate the effects of per-ceived friends' alcohol use, friends' alcohol attitudes, and knowledge of near-termhealth effects and alcohol prevalence on both contemporaneous and subsequentalcohol involvement. Evidence of mediation was obtained with both cross-sectionaland longitudinal findings. Perceived peer norms had a direct effect on alcohol use,and knowledge of normative alcohol use had a unique long-term protectiveinfluence on later alcohol use. Findings are discussed in terms of a 2-prongedprevention model that (a) integrates principles of social learning theory withexpectancy-based, cognitive-behavioral change and (b) emphasizes disseminationof age-appropriate alcohol information in programs that aim to reduce alcohol use.

A growing body of literature supports thestrength of alcohol-related expectancies asimportant determinants of adolescent drinkingbehavior (e.g., see Brown, Creamer, & Stetson,1987; Christiansen & Goldman, 1983; Gold-man, Brown, & Christiansen, 1987; Stacy,Widaman, & Marlatt, 1990). Studies of bothyouths and adults have demonstrated empiri-cally that perceived outcomes (i.e., anticipatedeffects) from drinking account for substantialamounts of variation in contemporaneous and

Lawrence M. Scheier and Gilbert J. Botvin,Department of Public Health, Institute for PreventionResearch, Cornell University Medical College.

Preparation of this article was partially supportedby a research grant (P50DA-7656) to Gilbert J.Botvin and a FIRST Award (R29-DA08909-01) toLawrence M. Scheier from the National Institute on DrugAbuse. Portions of these data were presented at themeeting of the Society for Prevention Research,Scottsdale, Arizona, June 1995.

Correspondence concerning this article should beaddressed to Lawrence M. Scheier, Department ofPublic Health, Institute for Prevention Research,Cornell University Medical College, 411 East 69thStreet, Kips Bay 201, New York, New York 10021.

longitudinal drinking patterns (Bauman & Che-noweth, 1984; lessor & lessor, 1977; Mosko-witz, Schaps, Schaeffer, & Malvin, 1984; Sher,Walitzer, Wood, & Brent, 1991; Stacy, New-comb, & Bentler, 1991) and predict well toproblem drinking among youth (Chen, Grube, &Madden, 1994; Christiansen, Smith, Roehling,& Goldman, 1989) and young adults (Brown,1985). Most of these studies have involved acentral theoretical premise that, in moderatedosages, alcohol has positive reinforcing proper-ties and that the learned contingency betweenalcohol and reinforcement or "expectance" islargely responsible for generating drinking behavior(Goldman & Rather, 1993; Lang & Michalec, 1990;Stacy, Leigh, & Weingardt, 1994).

In response to this growing body of knowl-edge, prevention programs have developedstrategies to deter youth from drinking thatfocus on changing beliefs regarding the benefi-cial effects of alcohol (e.g., Botvin & Botvin,1992). A central premise underlying theseinterventions is that effective barriers to alcoholconsumption include restructuring the cognitivelinkage between expectancy (i.e., attitude) andbehavior (e.g., Stacy, Bentler, & Flay, 1994).

48

Page 2: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

SOCIAL INFLUENCES, EXPECTANCIES, AND ALCOHOL USE 49

One means of changing behavior has been tooffer alternative and more enduring copingstrategies that obviate the need to dampen stresswith alcohol (Pentz, 1985). A short list of thesestrategies would include improved assertivenessskills (both social and drug refusal), self-reinforcement, and problem-solving confidence,all of which foster development of personalcompetence and self-efficacy skills. Use of thesetactics as part of mainstream prevention effortshas met with considerable success with bothnonclinical (Botvin, Baker, Dusenbury, Botvin,& Diaz, 1995) and clinical (Connors, Maisto, &Derman, 1992; Fromme, Kivlahan, & Marlatt,1986) populations.

Critical Nature of Adolescent Development

Existing models of expectancy theory andempirical findings from studies of adolescentalcohol use indicate that it is important forexaminations of the role of expectancies andtheir linkage to consumption to coincide withthe earliest stages of drinking. A majority ofyouth consume their first alcoholic beverageduring adolescence (Kandel, 1980; Newcomb &Bentler, 1986), and many youth will initiate alifelong history of drinking (Kandel & Logan,1984). Of particular importance is that mostadolescents drink with friends in a socialatmosphere and in moderate doses, leading tothe perception of immediate positive andsocially beneficial consequences. It seems onlynatural, then, that studies of the etiology ofadolescent alcohol use include a dual focus,incorporating memory-based or cognitive-basedprocesses (i.e., expectancies) along with a continuedemphasis on peer models for alcohol use.

Peer relations have long been implicated aspowerful causal agents in the determination ofalcohol use and are a central component of bothpeer cluster (Oetting & Beauvais, 1986) andself-derogation theories of adolescent alcoholand drug use (e.g., Kaplan, Martin, & Robbins,1984). In both approaches, peers represent amedium through which information regardingthe "beneficial" effects of alcohol is transferred,either vicariously (through observation) orthrough direct experience. For example, self-derogation theory suggests that deviant andalcohol-abusing peers offer positive reinforce-ment to already troubled and self-deprecating

youth, providing both sanctions for drinking andan emotional climate conducive to increasedself-efficacy (i.e., "If I drink I will be popularwith my friends"). According to this approach,alcohol consumption is driven in its earlieststages by a need for immediate social reward; atlater stages, alcohol becomes a palliative copingmechanism with heightened expectancies forboth social rewards and reduced negativeemotions (e.g., Cooper, Frone, Russell, &Mudar, 1995; Labouvie, 1986).

Despite the seemingly logical basis accordedto the relationship between peer influences and awide range of positive expectancies, few studieshave tested simultaneously the contribution ofsocial influences and expectancies to the predic-tion of alcohol. Equally compelling, even fewerstudies have elucidated the precise developmen-tal mechanisms that link these processes tosubsequent drinking. For instance, if peersrepresent a primary source of informationregarding the effects of alcohol (including bothknowledge of immediate consequences andknowledge pertaining to experiential facets ofalcohol's effects), then expectancies represent adirect by-product of these peer influences.Conceptually, then, because social facilitationexpectancies and their encoding into memoryare a natural consequence of observationallearning or direct modeling processes, andbecause social influences help to shape thedevelopment of expectancies, expectancies mayplay a mediational role in the prediction ofalcohol use. To test this hypothesis, the presentstudy used path-analytic techniques to examinethe direct and indirect relations of peer socialinfluences (both perceived behavior and atti-tudes), alcohol knowledge, and alcohol-relatedexpectancies for both concurrent and longitudi-nal alcohol involvement.

Conceptual Support for the MediationalRole of Expectancies

In recent years, several studies have sug-gested a possible chain of events that connectsboth social influences and expectancies to druguse (e.g., Abrams & Niaura, 1987; Bauman,Fisher, & Koch, 1989; Webb, Baer, Francis, &Caid, 1993). Most, if not all, of these studieshave purported that expectancies mediate a hostof risk factors for alcohol and other drugs.

Page 3: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

50 SCHEIER AND BOTVIN

According to Rohsenow, for example, "Drink-ers' expectancies about the consequences ofalcohol ingestion play a prominent role inmediating both alcohol ingestion and intoxi-cated behavior" (1983, p. 752). Webb et al.(1993) suggested that expectancies are the"cognitive channels through which importantsources of social influence . . . have theireffects" (p. 128). Goldman, Brown, Chris-tiansen, and Smith (1991) reaffirmed this viewwhen they stated that expectancies are part of acognitive enterprise representing memory (infor-mational) processes and that "memory pro-cesses should be examined as a possiblemediational mechanism" (p. 138). Also, accord-ing to Goldman et al. (1991), "alcohol expectan-cies are essentially concepts of if-then relation-ships between events or objects in the world andtheir consequences" (p. 139). Stacy et al. (1990)provided a reminder that, historically, "theconstruct of expectancy is hypothesized to be adominant and direct mediator of behavior" (p.918). Unfortunately, one or two factors may castdoubt on the strength of these claims. First,according to Leigh and Stacy (1991), manymodels of expectancy-behavior relations oftendo not meet statistical or empirical criteria formediation. Models that hypothesize expectan-cies as mediators of other key determinantsrequire specific tests of mediation (i.e., Baron &Kenny, 1986), and many researchers have oftennot met this challenge. A second concern is theabsence of compelling empirical evidence sup-porting a mediational role for expectancies(Bauman et al., 1989; Ennett & Bauman, 1991).

Empirical Validation of the MediationalRole of Expectancies

In some cases, researchers have relied onappropriate analytic strategies to test the postu-lated mediational role of expectancies; however,their results have not provided unequivocalsupport (e.g., Ennett & Bauman, 1991; Webb etal., 1993). Webb et al. (1993) examined therelative role of alcohol expectancies as media-tors of personality (sensation seeking andtolerance of deviance), peer influence (norma-tive expectations and attitudes toward alcohol),and parental attitudes toward alcohol use onsubsequent alcohol use in a cohort of adoles-cents. Their findings support the contention that

expectancies influence alcohol use; however,social and intrapersonal influences had substan-tially large direct effects on alcohol use indepen-dent of alcohol expectancies. All told, slightlyless than 25% of social risk and 29% ofintrapersonal risk were mediated by alcoholexpectancies. Thus, the notion that expectanciesare a final cognitive channel through which allsocial influences are processed was not entirelysupported.

Bauman and his colleagues have also exam-ined the intervening role of expectancies forboth alcohol (Ennett & Bauman, 1991) andtobacco use (Bauman et al., 1989). These studieshave included a wide array of peer and parentalsocial influence measures (perceived attitudesand normative expectations) hypothesized toinfluence intervening "expectancy" measures.Ennett and Bauman reported that perceived peerdrinking was significantly mediated throughpeer norms and several expectancy measures(social consequences and problem behaviorrelated to drinking). However, the introductionof the expectancy measures into the model didnot substantially reduce the direct effects of thepeer and parental attitudinal measures to alcoholuse, reinforcing the significant direct effectexerted by these risk factors.

Using prospective data, Bauman et al. (1989)found little empirical evidence to support thecontention that expectancies are a necessary andsufficient condition for mediation of social orpsychological (intrapersonal) influences on ciga-rette smoking or alcohol consumption (beer andhard liquor were analyzed separately). Notwith-standing these findings, other investigators haveprovided limited support for the mediationalrole of expectancies in studies of college-agedyouth and drinking (Henderson, Goldman,Coovert, & Carnevalla, 1994), drinking amonghigh-risk college-aged youth classified as chil-dren of alcoholics (Sher et al., 1991), andsubstance use among adolescents prospectivelyfollowed into young adulthood (Stacy et al.,1991).

Recently, Greenbaum, Brown, and Friedman(1995) reported that expectancies mediated 44%of the effect of delinquency and 31% of theeffect of antecedent oppositional symptoms onalcohol consumption among youth with conductdisorders (special education youth with emo-tional disturbances and youth in residential

Page 4: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

SOCIAL INFLUENCES, EXPECTANCIES, AND ALCOHOL USE 51

mental health treatment facilities). Despite thesupport for mediation proffered by this study,the select nature of the sample and the fact thatexpectancies were measured concurrently withthe criterion (and may have spuriously inflatedthe expectancy-drinking relationship) raise someconcern regarding the study's generalizabilityand predictive utility.

Consideration of the Roleof Knowledge Factors

Most current school-based drug abuse preven-tion programs include components that addressthe deleterious health effects, pharmacologiceffects, social liabilities, and legal risk related toconsumption of alcohol and other drugs (Botvin& Botvin, 1992; O'Connor & Saunders, 1992;Schaps, Moskowitz, Malvin, & Schaeffer, 1986).Current reviews, and even meta-analyses, havesystematically shown that alcohol and drugabuse prevention programs that depend onchanging knowledge structures as the primarymeans of altering behavior have fostered gainsin knowledge while not having an impact onbehavior. However, knowledge may play animportant role when considered in a multivariateframework along with other etiologic factors.For example, knowledge of the deleteriousconsequences of alcohol and other drug use mayact as a protective barrier and, despite strongsocial influences and peer pressure, attenuatesubstance use during the early experimentalstages. Yet, few models have tested the possibleinteraction of knowledge and expectancies inthe presence of social influences. In the currentstudy, we included two distinct measures ofalcohol knowledge: one tapping the perceivedeffects of alcohol and a second tapping knowl-edge regarding alcohol facts and perceivedprevalence of alcohol use. We expected thatknowledge would have a protective influenceand would reduce positive alcohol-related ex-pectancies and alcohol involvement. In sum, thecurrent study addressed some of the method-ological and conceptual problems in previousresearch by (a) using appropriate multivariatepath-analytic techniques (which correct for biasattributed to measurement error) to test themediational properties of social reinforcementexpectancies, (b) using both cross-sectional and3-year longitudinal data collected during a

critical period of development during whichexpectancies are likely to develop and be relatedto increased drinking behavior, and (c) includingsocial influences and measures of alcoholknowledge to ascertain empirically their predic-tive role as determinants of both expectanciesand later alcohol consumption.

Method

Sample

Data for the current study were obtained aspart of a 5-year investigation conducted between1985 and 1991 that was designed to study theetiology and prevention of tobacco, alcohol, andother illicit drug abuse. The study included 56public schools and was conducted at threesuburban sites, including central and easternupstate New York and Long Island. These rural,suburban, and urban locations are predomi-nantly (91%) White and middle class. Studentsin the seventh grade (Time 1: pretest; Time 2:3-month posttest), and annually thereafter (Time3-Time 6), were randomly administered threeforms of a closed-ended, group-administeredquestionnaire. Students were assured of theconfidentiality of their responses in writing(both on the parental consent form and on thequestionnaire itself), verbally at the time ofadministration, and through a certificate ofconfidentiality from the U.S. Department ofHealth and Human Services. Unique identifica-tion codes were litho-coded onto questionnairesand used to track students longitudinally. Itemsincluded in the survey assessed a variety ofpsychosocial, attitudinal, behavioral, and inter-personal items related to alcohol, tobacco, andmarijuana use.

The current analyses were based on a singleform that was randomly assigned to one third ofthe nontreatment students and that included therequisite measures of alcohol use, expectancies,social and normative influences, and alcoholknowledge. There were three waves of data forthe panel sample corresponding to the firstannual posttest administered in the 8th grade(Time 3) and two subsequent follow-up datacollections in the 9th (Time 4) and 10th (Time 5)grades. Two cohorts of data were available forthis study, one collected annually in the fall andone collected shortly thereafter in the spring.

Page 5: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

52 SCHEIER AND BOTVIN

Questionnaires were identical, and, for thepurposes of the current analyses, the cohortswere combined. Passive consent procedureswere used, and less than 1% of the total sampleincluded in the prevention trial refused participa-tion.

Alcohol, Social Influence, Expectancy,and Psychosocial Measures

Alcohol was measured identically at eachassessment. Measures included self-reportedfrequency of alcohol use ("How often, if ever,do you drink alcoholic beverages?"), quantity("How much, if at all, do you usually drink eachtime you drink?"), and drunkenness ("Howoften, if ever, do you get drunk?"). Additionaldrug use items assessed frequency of cigaretteuse ("How much do you generally smokenow?") and marijuana use ("How often [if ever]do you usually smoke marijuana?"), and weused these items in subsequent attrition analy-ses. Responses for the alcohol frequency itemranged from never tried them (1) to more thanonce a day (9); responses for the drinkingquantity item ranged from / don't drink (1) tomore than 6 drinks (6); and responses for thedrinking intensity item ranged from / don't drink(1) to more than once a day (9). For the purposeof conducting path analyses, we weighted andaveraged the three alcohol items using aproportional weighting scheme proposed byDouglass and Khavari (1982). Percentile-basedweighting effectively eliminates marked skew-ness (such nonnormality is often encounteredwith self-report drug use measures) and centersthe distribution on a midpoint corresponding tothe 50th percentile. Self-report measures ofalcohol consumption have been shown to bereliable and to provide accurate prevalenceestimates, particularly under conditions of ano-nymity and confidentiality (e.g., Gfroerer, 1985;Stacy, Widaman, Hays, & DiMatteo, 1985).

Two five-item scales were used to assessalcohol knowledge, one tapping knowledge ofthe effects of alcohol (e.g., "Alcohol tends topep a person up") and one tapping perceivedprevalence (e.g., "Most adults drink alcoholevery day") and health information (adversehealth effects) related to alcohol use (e.g., "Apregnant woman's drinking can affect the healthof her baby"). Scoring was based on dichoto-

mous true-false coding; false responses werecoded as 0 when incorrect and 1 when correct. Asingle item assessed perceived friends' attitudestoward alcohol ("How do your friends feelabout whether or not you drink?"), withresponses ranging from strongly against it (I) tostrongly in favor of it (5). A single item assessedperceived normative expectations for peer drink-ing ("In your opinion, how many people yourage drink alcoholic beverages?"), with re-sponses ranging from none (1) to almost all (6).A single item assessed perceived friends'alcohol use ("How many of your friends drinkalcohol?"), with responses ranging from none(1) to all or nearly all (5). Finally, a 7-item scaleassessed positive social reinforcement expectan-cies (e.g., "If kids drink alcohol, it provesthey're tough"; a = .88 at Time 3, a = .89 atTime 4, and a = .90 at Time 5). Responses forthe expectancy items ranged from stronglydisagree (1) to strongly agree (5).

Results

Attrition Analyses

Sample loss across all three waves of thestudy was similar to that of other large-scaleprevention studies (e.g., Hansen, Collins,Malotte, Johnson, & Fielding, 1985; Snow,Tebes, & Arthur, 1992). Sample loss wasprimarily due to absenteeism (including tru-ancy) and relocation to other school systems,although the field staff implemented an aggres-sive tracking procedure to locate students notpresent at the scheduled assessment (threemakeup testing sessions were provided forstudents not present at the regularly scheduledtesting session). Overall, 976 control studentswere available for cross-sectional analyses atTime 3 (eighth grade; 51.4% male), and 918students were available at Time 4 (51.9% male);the panel sample was composed of 789 students(Time 3-Time 5; 51.1% male; 19.2% loss fromTime 3 and 14.1% loss from Time 4). Therewere no significant differences in gender repre-sentation for the continuation and dropoutyouth.

Analyses of sample attrition comparing stu-dents who continued at each subsequent wave ofthe study (Times 3, 4, and 5) with those whoparticipated only at Time 3 revealed only a

Page 6: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

SOCIAL INFLUENCES, EXPECTANCIES, AND ALCOHOL USE 53

minimal effect of participant loss. Proportionalanalyses revealed a slight differential loss ofalcohol users among the continuing sample witheach follow-up wave. Between Time 3 and Time4, there was a significant loss of self-reporteddrinkers (71.48% of the panel sample and85.44% of dropouts reported using alcohol),X2(l, N = 976) = 9.10, p < .05; for the fullthree-wave panel sample, this differential losspersisted (70.47% of the panel sample and85.44% of dropouts), x20, N = 892) = 10.17,p<.001.

We also compared continuation and dropoutyouth on the complete set of social influence andexpectancy measures used in the longitudinalmodel. Additional psychosocial measures, in-cluded in the annual assessment as part of theintervention, were included in the attritionanalyses to determine whether there were anysystematic differences attributed to retentionstatus that would bias the panel sample.Psychometric properties and a complete exami-nation of the statistical relations of thesemeasures to alcohol and other drug use arecontained in Botvin (1993) and Scheier andBotvin(1995).

By comparison, students not present at allthree waves of the study reported lower levels ofalcohol knowledge (18.14 vs. 19.35), /(890) =2.58, p < .01, and perceived more friends asdrinking (2.44 vs. 2.18), f(890) = 2.07, p < .05.Dropouts also reported lower academic esteem(19.7 vs. 20.69), ?(856) = 2.22, p < .05, andmore alcohol-related deviance (on a five-itemunit-weighted index [i.e., "gotten into troublewith parents"]; 1.76 vs. 0.81), f(113) = 2.15,p < .05. The regression of attrition status(dropouts were coded as 0, and panel samplemembers were coded as 1) on the full set ofavailable baseline measures accounted for 5% ofthe variance and resulted in three significantpredictors: cigarette use (P = — .12, p < .05),deviance (P = -.10, p < .05), and marijuanause (3 = .10,p< .05).

The overall picture obtained from the attritionanalyses is that there were a few significant,albeit small, differences in psychosocial function-ing between panel and dropout students. More-over, as underscored by the regression analyses,these differences accounted for relatively smallamounts of variation in retention status. Thisfinding, coupled with the minimal behavioral

and psychosocial vulnerability accorded to thedropout students, suggested that we could stillexamine the utility and explanatory power of thelongitudinal model (i.e., external validity wouldbe minimally affected by the loss of higher endalcohol and drug users).

Gender Differences in Alcohol Use andPsychosocial Functioning

Means and standard deviations for male andfemale students present at each measurementpoint are provided in Table 1. The results ofthese between-groups (male vs. female) analy-ses underscore significant differences in thebehavioral measures and, likewise, in several ofthe exogenous social influence measures.1 Interms of the behavioral measures, male studentsdrank more frequently at Time 3, ?(953) =-3.02, p < .01, and reported a greater quantityof alcohol consumed, f(945) = 2.67, p < .01.There was a trend for male students to reportmore health information-prevalence knowl-edge, r(958) = 1.93, p < .06; more socialreinforcement expectancies, f(974) = 3.06, p <.01; and more positive attitudes of friendstoward drinking, f(974) = 2.72, p < .01. Femalestudents reported more perceived friends' drink-ing, f(974) = 2.36, p < .05, and greater peernorms for drinking, f(974) = 3.87, p < .001.

1 Despite the observation of slight behavioral andpsychosocial differences between male and femalestudents, the small sample sizes produced by separategender analyses and the difficulty of obtaining robustparameter estimation with small samples suggestedthat we analyze the combined data. Tanaka (1987)provided extensive documentation of the effect ofvariation in sample size on parameter estimationusing structural equations. Further elaboration of theproblems of sample variability and estimation proce-dures with structural equations, particularly the effectof sample size on goodness of fit indexes, has beenprovided by Marsh, Balla, and McDonald (1988). In arelated vein, MacKinnon, Warsi, and Dwyer (1995)conducted second-order Taylor series estimationswith simulated data (N = 500 replications) andreported that a sample size of 500 was sufficient foraccurate point and variance estimates of the propor-tion mediated (using a dichotomous predictor withcontinuous mediator and outcome measures). Unfor-tunately, conducting analyses separately for each sexgroup would not have met this statistical criterion.

Page 7: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

54 SCHEIER AND BOTVIN

>o^J

.1ca

1.

•1

a%>

1

1eoU

•2

Q-g53•§ato"ia

EP

<uEP

CO

s,p

U £

l§1> &fc %

a> S

^ 3M

u S

E -a£ 1

in

ju c

u 42•a gE-S£ s

VI

^ cWJ -rt

«

^Oo

oo ^o o*jON fS OO•* —; o

i O roi (N vd

Tt •— • m O '-*

S >n — <r -~ooc

moo — <N — — en o ON

oooooio o o ^ ^ ^ - - »

>.g .a

~

o

•a

fa O

o '5^^ e

lliS 2o H

At each successive follow-up wave, severaldifferences in the behavioral, social influence,and psychosocial measures persisted. At the firstfollow-up (Time 4), male students continued toreport drinking more frequently, f(851) = 3.03,p < .01; greater quantities of alcohol consumed,f(857) = 2.44, p < .05; and more drunkenness,f(810) = 2.28, p < .05. Reversing an earliermarginal trend, female students reported moreprevalence knowledge, f(855) = 4.71, p < .001,and greater peer normative expectations fordrinking, f(866) = 3.36,p < .001, whereas malestudents reported more favorable attitudes offriends toward drinking, f(863) = 2.86, p < .01.Most of these differences continued unabated,and, for the full panel sample, male studentsreported drinking more frequently, f(793) =2.39, p < .05; greater quantity consumed,f(806) = 2.23, p < .05; more drunkenness,r(779) = 2.11,p < .05; less health information-prevalence knowledge, ?(775) = 5.31,p < .001;and greater social reinforcement expectancies,r(794) = 4.89, p < .001. There was also amarginal trend for male students to report morepositive attitudes of their friends toward alcoholuse, f(798) = 1.94, p < .06. In contrast, femalestudents reported greater levels of peer norma-tive expectations for alcohol, f(782) = 4.75,p<.001.

In terms of the psychosocial measures at Time3, female students also reported lower assertive-ness skills, lower self-esteem, less deviance,more decision-making skills, higher locus ofcontrol (scored toward externality), higherproblem-solving confidence, more depressiveand anxious symptoms, and more social concern(ps < .01), as well as marginally higher levelsof social confrontation and self-managementskills (ps < .06). With few exceptions, thepatterns and magnitudes of these differencespersisted across subsequent assessments, indicat-ing that attrition did not systematically bias thesample. Female students reported significantlymore behavioral control at Time 4 than at Time3. Gender differences for the complete mergedpanel sample that differed from those alreadyreported included more self-management skillsfor female students (p < .001), as well as lessexternal locus of control (p < .05).

In addition to gender differences across thethree waves, within-group (male students andfemale students) across-time differences were

Page 8: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

SOCIAL INFLUENCES, EXPECTANCIES, AND ALCOHOL USE 55

also noted (see Table 1). Paired t tests were usedin these analyses to account for the dependencein scores across each wave of measurement. Forthe Time 3-Time 4 analyses, male studentsshowed increases in terms of drinking fre-quency, f(449) = 5.78, p < .001; quantity,f(449) = 7.99, p < .001; self-reported drunken-ness, f(449) = 5.90, p < .001; social reinforce-ment expectancies, r(449) = 3.71, p < .001;perceived friends' alcohol use, f(449) = 9.59,p < .001; positive attitudes of friends towardalcohol, f(449) = 4.45, p < .001; and perceivedpeer normative alcohol use, t(449) = 6.83, p <.001. Female students showed increases in termsof frequency of drinking, f(424) = 7.24, p <.001; quantity, f(424) = 10.26, p < .001;drunkenness, f(424) = 7.12, p < .001; healthinformation-prevalence knowledge, ?(424) =1.97, p < .05; social reinforcement expectan-cies, f(424) = 2.83, p < .01; perceived friends'alcohol use, ?(424) = 9.11, p < .001; positiveattitudes of friends toward alcohol, f(424) =5.20, p < .001; and perceived peer normativealcohol use, f(424) = 6.93, p < .001.

The patterns of increased alcohol use, quan-tity, and drunkenness persisted for male studentscontinuing in the panel sample (ps < .001), asdid their increased health information-preva-lence knowledge, f(403) = 2.34, p < .05;perceived friends' alcohol use, f(403) = 8.73,p < .001; positive attitudes of friends towardalcohol, f(403) = 2.81, p < .01; and peernormative expectations, r(403) = 5.50, p <.001. Female students continuing in the panelsample also showed increased levels of drink-ing, quantity, and drunkenness (ps < .001), aswell as health information-prevalence knowl-edge, f(386) = 3.70, p < .001; perceivedfriends' alcohol use, f(386) = 9.51, p < .001;positive attitudes of friends toward alcohol,f(386) = 4.00, p < .001; and peer alcoholnorms, f(386) = 8.01, p < .001.

Prevalence of Alcohol Use andTransitions in Use

At baseline (Time 3), 73% of the controlsample reported some use of alcohol. Amongthese nonabstaining youth, 50% reported drink-ing alcohol within the past month, slightly lessthan 25% reported doing so within the pastweek, and 7% reported doing so the day before

the assessment. Prevalence of alcohol useincreased over the 3-year period, and, at the9th-grade assessment, 82% of the studentsreported using alcohol. Of these nonabstainingyouth, 58% reporting drinking in the pastmonth, 30% reported drinking in the past week,and 11% reported drinking the prior day. At thelOth-grade assessment, 87.4% of the continuingstudents reported use of alcohol, 63% in the pastmonth, 34% in the past week, and 10% on theday before. Proportional analyses revealedsignificant increases in the number of alcoholusers between Time 3 and Time 4, x20> N =

873) = 230.23, p < .001 (51.5% new users),and also for the longitudinal panel sample(Times 3-5, including all students present atboth Time 3 and Time 4), X

2(l, N = 823) =131.50, p < .001 (66.7% new users). Theincreased alcohol use patterns were evident forboth male and female students (ps < .001).

Cross-Sectional Mediational Model

We conducted tests of the mediational proper-ties of expectancies using the EQS structuralequation modeling program (Bentler, 1989).Statistical conventions available in EQS allowfor decomposition of effects (direct and indirect)when mediational models are posited. Baronand Kenny (1986) suggested path analysis as anappropriate analytic framework for testingmediational models; however, the current modelcontained multiple exogenous measures, compli-cating the methods for testing the significance ofmediation. As one of many statistical features,the EQS program extended the work of Sobel(1987) by providing an empirical determinationof whether indirect effects are significantlydifferent from zero (based on the z-critical ratioof the nonstandard effect to its standard errorestimate). A variance-covariance matrix wasinput for all subsequent modeling.

Figure 1 contains the standardized solutionfor the Time 3 cross-sectional model (i.e., modelvariables rescaled to have unit variance). Thenumbers inside the circles designate residualvariances (i.e., net predictive variance unac-counted for by the exogenous variables). Asdepicted, the model fully explains 49% of thevariance (1.00 — .51) in alcohol consumption(and adjusts down to 48.6% when all nonsignifi-cant paths are removed). The fit of this final

Page 9: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

56 SCHEIER AND BOTVIN

KnowledgeEffects

KnowledgeFacts/ Prevalence

Peer NormativeExpectations

Friends' Attitudeto Alcohol

Perceived Friends'Alcohol Use

XAlcohol

Expectancies

®/

.22*" AlcoholUse

r.08*

.15*

.42*

Figure /. Results of cross-sectional model depicting relations between social influences, expectancies, andalcohol use. AH variables are measured, and small circles contain residual variances (net after prediction).Significances are determined by ratio of nonstandard parameter estimate divided by its standard error. (**p <.01 ;***/><. 001).

model (with nonsignificant paths removed) wasadequate, x2(4, N = 823) = 2.98, p = .56,comparative fit index (CFI) = 1.0. The CFI is anoverall index of fit that determines the extent towhich sample variances and covariances arereproduced by the hypothesized model struc-ture. This statistic ranges from 0 to 1; bench-marks approaching .90 are indicative of a goodfit (Bender, 1990). The ratio of chi-square valueto degrees of freedom was well within the 5:1ratio suggested by Bollen (1989), providinganother indicator of the adequate fit of thismodel.

As depicted in Figure 1, only three measuresdirectly influenced contemporaneous alcoholuse: perceived friends' alcohol use (|3 = .42),friends' attitudes toward alcohol ((3 = .15), andpeer normative expectations ((3 = .08). Theeffects of these measures, other than peernormative expectations, were significantly medi-ated through expectancies, with the largest ofthese effects attributed to perceived friends'alcohol use (f3 - .31). Both knowledge mea-sures were inversely associated with the expec-

tancy measure, reinforcing the protective effectsof factual information and knowledge regardingnegative effects of alcohol consumption.2

Factor intercorrelations among the exogenoussocial influence measures for the cross-sectionalmodel are included in Table 2 (and should beread in combination with Figure 1). As indi-cated, with the exception of knowledge of theeffects of alcohol and friends' attitudes towardalcohol, associations between the exogenousmeasures were significant and in the hypoth-esized direction. The association between healthinformation-prevalence knowledge and knowl-edge of the effects of alcohol was inverse andsmall, albeit significant, underscoring the distinc-tive nature of alcohol information. The correla-tive patterns between each of the knowledge

2 There were direct effects for the exogenousmeasures on alcohol use lhat did not involveestimation of the indirect path from expectancies toalcohol use and that differed from the cross-sectionalmodel shown only for health information-prevalenceknowledge (P = -.06, p < .05).

Page 10: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

SOCIAL INFLUENCES, EXPECTANCIES, AND ALCOHOL USE 57

Table 2Correlations Among Exogenous Measures: Cross-Sectional Model

Variable 1

1 . Knowledge of effects2. Health information-prevalence knowledge3. Friends' alcohol use4. Friends' alcohol attitudes5. Peer normative expectations

-.07*-.05*

nsa

-.06*

-.19***-.14***-.13***

.53*** —

.54*** .37*** —

Note. All significant correlations were determined by z-critical tests (one-tailed z = 1.64)."Constrained to zero in the final path model.

scales and the remaining social influencemeasures also underscore the conceptual diver-gence between these two scales. Associationsbetween health information-prevalence knowl-edge and social influences were of a largermagnitude than the comparable associationsbetween knowledge of the effects of alcohol andsocial influences. Among the social influencemeasures, perceived friends' alcohol use, friends'attitudes toward drinking, and peer norms foralcohol use were moderately and significantlyassociated. Finally, we used MacKinnon's (1994)formula for determining the proportion medi-ated in each of the models tested. The percent-age of the total effect that was mediated wascomputed only for cases in which there was anonzero direct path and a significant indirectpath (MacKinnon, Warsi, & Dwyer, 1995). Theresulting percentages were 14% for perceivedfriends' alcohol use and 26.8% for friends'attitudes toward alcohol.

Longitudinal Path Analyses

The next step in the analyses included testinga three-wave longitudinal model with alcoholexpectancies hypothesized as a mediating vari-able. We configured this model with three wavesof data to permit the simultaneous testing of thegeneration of expectancies (from social influ-ences), the long-term prediction of alcoholinvolvement from social influences, and theprediction, over 1 year, of alcohol use fromexpectancies. In this manner, we were able totest the mechanism by which expectancies"filter" the effects of social influences and drugknowledge on later alcohol use. Figure 2contains the results of the longitudinal three-wave path model (nonsignificant paths were

removed). In this model, the exogenous mea-sures correspond to the Time 3 assessment, theexpectancies correspond to the Time 4 assess-ment, and alcohol consumption corresponds tothe Time 5 assessment. To control for spuriousrelations, we also included an earlier measure ofthe consequent (alcohol involvement) at base-line. All of the exogenous measures wereallowed to freely covary, primarily becausespecifications of causal relations among contem-poraneous measures are at best tenable (thesecorrelations are included in Table 3).

With few exceptions, results of the longitudi-nal model replicated the cross-sectional find-ings, although the diminution of effect sizessuggests some temporal erosion. Fit indexes forthis model indicated an adequate fit, x2(6) =6.22, p = .40, CFI = 1.0, \2:df < 1.0, andreinforced that we correctly hypothesized thestructural relations among the measures. Asdepicted, alcohol involvement was moderatelystable over the 3-year period ((3 = .43), andearly-stage alcohol use indirectly influencedalcohol expectancies (P = .27). Again, bothknowledge measures were inversely related tothe expectancies measure, with health informa-tion-prevalence knowledge more strongly corre-lated than knowledge of the effects of alcohol.Health information-prevalence knowledge hada small but significant long-term protectiveeffect on later alcohol use ((3 = -.05). With theexception of peer norms, which had no signifi-cant indirect or direct effects, expectanciessignificantly mediated the effects of all of theremaining social influence measures. Perceivedfriends' alcohol use, but not friends' attitudestoward alcohol use, also increased later alcoholconsumption (P = .09). Finally, expectancies

Page 11: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

58

T3 (8th Grade)

SCHEIER AND BOTVIN

T4 (9th Grade) T5 (10th Grade)

AlcoholUse

KnowledgeEffects

KnowledgeFacts/ Prevalence

Friends' Attitudeto Alcohol

Perceived Friends'Alcohol Use

.43*

-.17*

-^^Alcohol

Expectancies

/ X•08* ^ Alcohol

H Use

Peer NormativeExpectations

Figure 2. Longitudinal path model depicting mediational properties of expectancies (*p < .05; ***/> < .001).T3 = Time 3; T4 = Time 4; T5 = Time 5.

significantly predicted alcohol use over the1-year period ((3 = .08).3

Again, we computed the percentage media-tion for the longitudinal model for all conditionsin which there was a nonzero direct effect(MacKinnon et ah, 1995). Expectancies medi-ated 4.8% of Time 3 alcohol use, 21.4% ofprevalence knowledge, and 11.1% of perceivedfriends' alcohol use. Despite these relativelysmall mediated proportions, it is important toremember that the remaining exogenous mea-sures had no significant direct effects on thecriterion, and all were indirectly channeledthrough expectancies, reinforcing the strength ofexpectancies as an intervening mechanism.

Discussion

During the early stages of adolescence,alcohol consumption is developmental Iy tied tosocial reward systems and peer influences.Because perhaps the single most importantreason adolescents drink is the need for social

approval, and because this approval is largelyconveyed by their immediate peers, it becomesimportant to investigate the developmentalmanner in which social learning factors influ-ence cognitive memory systems in a singlemodel. In addition to their immediate role aspart of the social influence process, peers alsoconvey substantial knowledge regarding theeffects of alcohol and represent an importantsource of factual information regarding near-term health effects of alcohol. In the currentstudy, we captured the temporal and logicalorder of these relationships and, using amultivariate framework, tested whether alcohol-related social reinforcement expectancies would

(There were no effects in the modified model(constraining the indirect path from Time 4 expec-tancy to Time 5 alcohol use) that differed from thelongitudinal model depicted in Figure 2 (the effectsizes for existing paths differed slightly in magni-tude).

Page 12: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

SOCIAL INFLUENCES, EXPECTANCIES, AND ALCOHOL USE 59

Table 3Correlations Among Exogenous Measures: Longitudinal Path Model

Variable 1

1 . Alcohol use2. Knowledge of effects -.06*3. Health information-prevalence knowledge -.18*** nsa —4. Friends' alcohol use5. Friends' alcohol attitudes6. Peer normative expectations

.65***

.51***

.41***

-.05*ns*

-.06*

-.17***-.10**-.11**

—.53*** —.55*** .40*** —

"Constrained to zero in the final path model.*p < .05. **p < .01. ***p < .001 (one-tailed, z = 1.64).

mediate the effects of social influences andalcohol knowledge on both concurrent andsubsequent alcohol use.

The results of this study confirm thatexpectancies play an important and key interven-ing role in the promotion of adolescent alcoholuse. In the cross-sectional model, four of the fivesocial influence measures had significant indi-rect effects on alcohol expectancies, and two ofthese measures, perceived friends' alcohol useand friends' attitudes toward alcohol use, hadsignificant direct effects on alcohol consump-tion. Likewise, peer normative expectations hadonly a contemporaneous direct effect on alcoholuse, suggesting that during this early part ofadolescent development perception of peernorms for alcohol use was not an essentialingredient in the construction of alcohol-relatedexpectancies.

In the longitudinal model, four of the fivesocial influence measures were significantlymediated through expectancies (but not norma-tive expectations) after control for early levelsof alcohol use, reinforcing the importance ofinternal cognitive processes in the early stagesof alcohol use. In addition to the indirect paths,perceived friends' alcohol use had a small butsignificant direct effect on later alcohol use. Netof the effect of social influences, expectanciesalso significantly predicted later drinking. Thesefindings paint a more detailed picture regardingthe cognitive processes by which peer socialrelations influence the development of bothexpectancies and later alcohol use. What thesefindings highlight is that both early-stage andlater (and more pronounced) drinking arestrongly predicated on social influences but thatlater drinking is also instrumentally tied to

cognitive memory processes that reflect internalreward systems.

In contrast to the findings of this study, Ennettand Bauman (1991) reported that social influ-ences still maintained a relatively large directeffect on alcohol use even when expectancieswere specified in the same model (as mediators).Because they included parental alcohol use as ameasure of social influence, and because manyyouth may not vicariously observe their parents'drinking behavior, this may prevent directcomparison with the current findings. Ourresults point more conclusively toward thenecessity of a cognitive mechanism in theelaboration of social influence processes andalcohol use. Given the absence of convergentfindings between these studies, further researchis warranted that can replicate the currentfindings and clarify the mediational role ofexpectancies.

The current study also extended the timeinterval over which the effects of socialinfluences and expectancies on alcohol use wereexamined, a period corresponding to the criticalyears during which adolescents are most likelyto initiate and consolidate these behaviors. Thisis an important component of developmentalresearch and highlights the importance ofvarying time intervals to establish empiricallywhether causal relations are bound by time. Thefindings from the cross-sectional model wereextended by the longitudinal model, whichspanned 3 years, and underscored the findingthat the perception of immediate social facilita-tion from drinking remains predictive of alcoholuse 1 year later. The increased mean levels ofself-reported drinking in this sample alsosupport the increasing variability and growing

Page 13: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

60 SCHEIER AND BOTVIN

consolidation of behavior over time. In 10th-grade self-reports, as compared with Sth-gradereports, more youth had experienced alcohol,and mean levels of alcohol use (includingdrunkenness) were significantly elevated forboth male and female students. These increaseswere not limited to behavioral measures but alsoincluded elevated levels of alcohol expectanciesand social influences.

Limitations

There are several limitations to the currentstudy worth noting. First, the sample weanalyzed represents a broad mixture of experi-mental alcohol use patterns, including youthreporting no alcohol use and youth reportingfrequent (if not abusive) use. As others havenoted, differences in relevant alcohol experi-ences can moderate the expectancy-behaviorrelationship (Cooper et al., 1995), and thisrelationship may also be reciprocal (e.g., Smith,Goldman, Greenbaum, & Christiansen, 1995).In a related vein, path analysis is helpful inelucidating the simultaneous relations amongmultiple indicators of social influence andproviding estimates for unique effects over time.The processes that are highlighted by the pathmodel reflect group-level, rather than individual-level, behavioral tendencies, and future analysesmay want to examine individual variation inexpectancies (high vs. low) or alcohol use(nonuse vs. use) and the manner in which theserisk mechanisms influence future consumption.

Second, the expectancies we examined in thecurrent study tap perceived social facilitationand do not emphasize physical or cognitiveexpectancies (i.e., tension reduction), the latterof which are also influenced by social andenvironmental cues (e.g., Christiansen, Gold-man, & Inn, 1982; Fromme & Dunn, 1992).Thus, future research may want to draw from abroader pool of expectancy items and determinewhether the current findings are specific tosocial reinforcement motivations for alcohol useat this early age. It is quite possible, however,that in the early stages of alcohol use, socialreinforcement expectancies are prominent andthat with continued and exacerbated alcohol use,linkages of behavior to physiological arousal,tension reduction, and emotional regulationbecome more salient (e.g., Labouvie, 1986).

Related to this problem is that our modelexcluded any consideration of the role ofnegative expectancies. In light of the biphasicproperties of alcohol and the immediacy ofpositive responses to drinking (i.e., fosteringenhanced retrieval), we are inclined to support amodel that hypothesizes a dominance of positivesocial reinforcement expectancies at the earlieststages of drinking. However, Stacy and hiscolleagues have provided both conceptual argu-ments and empirical support for the inclusion ofnegative expectancies, which, among older andmore experienced drinking populations, may beequally potent in predicting consumption (Stacy,MacKinnon, & Pentz, 1993; Stacy, Newcomb,& Bentler, 1991) and should be included inmodels that seek to elaborate a more completeunderstanding of expectancy-behavior rela-tions.

Third, in the interest of parsimony, theexternal variables in the longitudinal modelwere limited to those tapping social influences,which represent only a limited set of the totalarray of psychosocial influences that may affectalcohol use (e.g., Hawkins, Catalano, & Miller,1992). Among the many indexes of model fit,the overall R2 values for both the cross-sectionaland longitudinal models underscore that alterna-tive models with increased numbers of predic-tors might enhance prediction (48% in thecross-sectional model and 31% in the longitudi-nal model). Future studies that specify expectan-cies as mediators of psychosocial influencesmay want to include additional personality,intrapersonal, demographic, and interpersonalmeasures that are part of the network ofinfluences causally related to alcohol use.Unfortunately, we were unable to include manyof these measures, principally because robustestimation with path analysis is severely con-strained with moderately small sample sizes(Tanaka, 1987). Expanding the size of ourmodel to include elements of any of theseimportant domains of risk might have beenconceptually accurate; however, it would havestrained the robustness of the analytic methodswe used to examine the data.

Implications for Prevention

This study failed to find any consistent patternof relations among peer normative expectations,

Page 14: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

SOCIAL INFLUENCES, EXPECTANCIES, AND ALCOHOL USE 61

alcohol expectancies, and alcohol use. In thecross-sectional model, normative expectationshad a small but significant effect on alcohol use,whereas, in the longitudinal model, the samemeasure did not significantly influence eitherexpectancies or later alcohol use. Peer norma-tive expectations for alcohol use were moder-ately related to friends' alcohol use and alcoholattitudes and less related to health information-prevalence knowledge. Because one of thestrengths of path analysis is that it providesunique parameter estimates, it is possible thatthe moderate overlap between peer norms andpeer models for drinking prevents both of thesemeasures from uniquely predicting expectanciesand consumption. The absence of unique effectsassociated with peer norms does not diminishthe importance of correcting erroneous norma-tive expectations regarding both peer and adultalcohol use and maintaining this key focus ofprevention (Botvin & Botvin, 1992; Graham,Marks, & Hansen, 1991).

It is also important to understand that peerswill always exert a strong developmentalinfluence, particularly because peer social rela-tions are essentially the means by which youthtest, establish, and refine their psychologicalidentity (e.g., Seltzer, 1989). Ideally, peerrelations should represent a positive medium foridentity development; in certain situations,however, peers may represent a safe haven forfallout from family problems and personalidentity crises and may provide opportunities forescapism through excessive alcohol and druguse. In these extreme situations, expectanciesmay represent a microscope that helps focus onthe source of discontent beneath the emotion-focused coping that alcohol represents.

Finally, the consistent negative pattern ofrelations among both facets of alcohol knowl-edge (effects and health information-preva-lence), expectancies, and social influencesrepresents an important vehicle for the effectiveconductance of prevention. Factual informationremains an effective deterrent against alcoholuse and should be included as a component ofprevention programs. Botvin (1995) has arguedthat, to be effective, information-based preven-tion programs need to emphasize the adverseconsequences of drug use, and informationshould be developmentally appropriate andconsider the needs of youth (specifically their

rational capabilities and personal interests).Included in these recommendations is a concernthat adolescents may not respond to specificexhortations regarding drug abuse as addictive,particularly because many young people main-tain an illusion of invulnerability. In light ofthese concerns, the measures of alcohol knowl-edge that we included assess near-term healthconsequences (e.g., "Switching drinks willmake you drunker than staying with the samekind of alcoholic beverage") and health informa-tion-prevalence (e.g., "Alcohol is the cause ofthe majority of fatal car accidents"), measuresthat should have immediate relevance to adoles-cent concerns. Health information-prevalenceknowledge was a stronger predictor of expectan-cies and, in the longitudinal model, had a smallbut significant negative effect on later alcoholuse. Accordingly, exclusion of alcohol informa-tion from current prevention efforts on the solebasis that previous information-based preven-tion approaches have not been able to showmarked reductions in target behaviors or demon-strate significant statistical relationships be-tween knowledge and alcohol use may beerroneous (e.g., Schaps, Moskowitz, Condon, &Malvin, 1982). It is quite possible that theinformation component of prevention programsserves as a foundation to the "cognitive" facetof attitude formation and that, if there issufficient time for a linking with other develop-mental facets of expectancies, greater successwill be achieved in deterring alcohol use. Ingeneral, expectancies may represent a dynamicwindow through which a confluence of socialand intrapersonal forces actively operate topromote a variety of health-related and health-compromising behaviors.

References

Abrams, D. B., & Niaura, R. S. (1987). Sociallearning theory. In H. T. Blane & K. E. Leonard(Eds.), Psychological theories of drinking andalcoholism (pp. 131-178). New York: GuilfordPress.

Baron, R. M., & Kenny, D. A. (1986). Themoderator-mediator variable distinction in socialpsychological research: Conceptual, strategic, andstatistical considerations. Journal of Personalityand Social Psychology, 51, 1173-1181.

Bauman, K. E., & Chenoweth, R. L. (1984). Therelationship between the consequences adolescents

Page 15: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

62 SCHEIER AND BOTVIN

expect from smoking and their behavior: A factoranalysis with panel data. Journal of Applied SocialPsychology, 14,1%-^\.

Bauman, K. E., Fisher, L. A., & Koch, G. G. (1989).External variables, subjective expected utility, andadolescent behavior with alcohol and cigarettes.Journal of Applied Social Psychology, 19, 789-804.

Bentler, P. M. (1989). EQS structural equationsprogram manual. Los Angeles: BMDP StatisticalSoftware.

Bentler, P. M. (1990). Comparative fit indexes instructural models. Psychological Bulletin, 107,238-246.

Bollen, K. A. (1989). Structural equations with latentvariables. New York: Wiley.

Botvin, G. J. (1993). Reducing drug abuse and AIDSrisk: Final report. Rockville, MD: National Instituteon Drug Abuse.

Botvin, G. J. (1995). Drug abuse prevention in schoolsettings. In G. J. Botvin & M. A. Orlandi (Eds.),Drug abuse prevention with multiethnic youth (pp.169-192). Thousand Oaks, CA: Sage.

Botvin, G. J., Baker, E., Dusenbury, L., Botvin, E., &Diaz, T. (1995). Long-term follow-up of a random-ized drug abuse prevention trial in a Whitemiddle-class population. Journal of the AmericanMedical Association, 273, 1106-1112.

Botvin, G. J., & Botvin, E. M. (1992). School-basedand community-based prevention approaches. In J.H. Lowinson, P. Ruiz, & R. B. Millman (Eds.),Substance abuse: A comprehensive textbook (2nded., pp. 910-927). Baltimore: Williams & Wilkins.

Brown, S. A. (1985). Expectancies versus back-ground in the prediction of college drinkingpatterns. Journal of Consulting and Clinical Psychol-ogy, 53, 123-130.

Brown, S. A., Creamer, V. A., & Stetson, B. A. (1987).Adolescent alcohol expectancies in relation topersonal and parental drinking patterns. Journal ofAbnormal Psychology, 96, 117-121.

Chen, M.-J., Grube, J. W., & Madden, P. A. (1994).Alcohol expectancies and adolescent drinking:Differential prediction of frequency, quantity, andintoxication. Addictive Behaviors, 19, 521-529.

Christiansen, B. A., & Goldman, M. S. (1983).Alcohol-related expectancies versus demographic/background variables in the prediction of adolescentdrinking. Journal of Consulting and ClinicalPsychology, 51, 249-257.

Christiansen, B. A., Goldman, M. S., & Inn, A.(1982). Development of alcohol-related expectan-cies in adolescents: Separating pharmacologicalfrom social-learning influences. Journal of Consult-ing and Clinical Psychology, 50, 336-344.

Christiansen, B. A., Smith, G. T., Roehling, P. V., &Goldman, M. S. (1989). Using alcohol expectanciesto predict adolescent drinking behavior after one

year. Journal of Consulting and Clinical Psychol-ogy, 57, 93-99.

Connors, G. J., Maisto, S. A., & Derman, K. H.(1992). Alcohol-related expectancies and theirapplications to treatment. In R. R. Watson (Ed.),Drug and alcohol abuse reviews: Vol. 3. Alcoholabuse treatment (pp. 203-231). Totowa, NJ: Hu-mana Press.

Cooper, M. L., Frone, M. R., Russell, M., & Mudar, P.(1995). Drinking to regulate positive and negativeemotions: A motivational model of alcohol use.Journal of Personality and Social Psychology, 69,990-1005.

Douglass, F. M., & Khavari, K. A. (1982). A majorlimitation of the percentile index of overall drug useindulgence. International Journal of the Addictions,17, 283-294.

Ennett, S. T., & Bauman, K. E. (1991). Mediators inthe relationship between parental and peer character-istics and beer drinking by early adolescents.Journal of Applied Social Psychology, 21, 1699-1711.

Fromme, K., & Dunn, M. E. (1992). Alcoholexpectancies, social and environmental cues asdeterminants of drinking and perceived reinforce-ment. Addictive Behaviors, 17, 167-177.

Fromme, K., Kivlahan, D. R., & Marlatt, G. A.(1986). Alcohol expectancies, risk identification,and secondary prevention with problem drinkers.Advances in Behavioral Research and Therapy, 8,237-251.

Gfroerer, J. (1985). Influence of privacy on self-reported drug use by youths. In B. A. Rouse, N. J.Kozel, & L. G. Richards (Eds.), Self-report methodsof estimating drug use: Current challenges tovalidity (DHHS Publication No. ADM 85-1402, pp.22-30). Washington, DC: U.S. Government Print-ing Office.

Goldman, M. S., Brown, S. A., & Christiansen, B. A.(1987). Expectancy theory: Thinking about drink-ing. In H. T. Blane & K. E. Leonard (Eds.),Psychological theories of drinking and alcoholism(pp. 181-226). New York: Guilford Press.

Goldman, M. S., Brown, S. A., Christiansen, B. A., &Smith, G. T. (1991). Alcoholism and memory:Broadening the scope of alcohol-expectancy re-search. Psychological Bulletin, 110, 137-146.

Goldman, M. S., & Rather, B. C. (1993). Substanceabuse: Cognitive models and architecture. In P. C.Kendall & K. Dobson (Eds.), Psychopathology andcognition (pp. 245-292). New York: AcademicPress.

Graham, J. W., Marks, G., & Hansen, W. B. (1991).Social influence processes affecting adolescentsubstance use. Journal of Applied Psychology, 76,291-298.

Greenbaum, P. E., Brown, E. C., & Friedman, R. M.

Page 16: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

SOCIAL INFLUENCES, EXPECTANCIES, AND ALCOHOL USE 63

(1995). Alcohol expectancies among adolescentswith conduct disorder: Prediction and mediation ofdrinking. Addictive Behaviors, 20, 321-333.

Hansen, W. B., Collins, L. M., Malotte, C. K.,Johnson, C. A., & Fielding, J. E. (1985). Attrition inprevention research. Journal of Behavioral Medi-cine, 8, 261-275.

Hawkins, J. D., Catalano, R. F., & Miller, J. Y. (1992).Risk and protective factors for alcohol and otherdrug problems in adolescence and early adulthood:Implications for substance abuse prevention. Psycho-logical Bulletin, 112, 64-105.

Henderson, M. J., Goldman, M. S., Coovert, M. D., &Carnevalla, N. (1994). Covariance structure modelsof expectancy. Journal of Studies on Alcohol, 55,315-326.

lessor, R., & lessor, S. L. (1977). Problem behaviorandpsychosocial development: A longitudinal studyof youth. New York: Academic Press.

Kandel, D. B. (1980). Drug and drinking behavioramong youth. Annual Review of Sociology, 6,235-285.

Kandel, D. B., & Logan, J. A. (1984). Patterns of druguse from adolescence to young adulthood: I. Periodsof risk for initiation, continued use, and discontinu-ation. American Journal of Public Health, 74,660-666.

Kaplan, H. B., Martin, S. S., & Robbins, C. (1984).Pathways to adolescent drug use: Self-derogation,peer influence, weakening of social controls, andearly substance use. Journal of Health and SocialBehavior, 25, 270-289.

Labouvie, E. (1986). The coping function of adoles-cent alcohol and drug use. In R. K. Silbereisen, K.Eyferth, & G. Rudinger (Eds.), Development asaction in context (pp. 229-240). New York:Springer-Verlag.

Lang, A. R., & Michalec, E. M. (1990). Expectancyeffects in reinforcement from alcohol. In W. M. Cox(Ed.), Why people drink: Parameters of alcohol as areinforcer (pp. 193-232). New York: GuilfordPress.

Leigh, B. C., & Stacy, A. W. (1991). On the scope ofalcohol expectancy research: Remaining issues ofmeasurement and meaning. Psychological Bulletin,110, 147-154.

MacKinnon, D. P. (1994). Analysis of mediatingvariables in prevention and intervention research. InA. Cazares & L. A. Beatty (Eds.), Scientific methodsfor prevention intervention research (NIDA Re-search Monograph 139, NIH Publication No. ADM94-3631, pp. 127-153). Washington, DC: U.S.Government Printing Office.

MacKinnon, D. P., Warsi, G., & Dwyer, J. H. (1995).A simulation study of mediated effect measures.Multivariate Behavioral Research, 30, 41-62.

Marsh, H. W., Balla, J. W., & McDonald, R. P. (1988).

Goodness-of-fit indices in confirmatory factoranalysis: Effects of sample size. PsychologicalBulletin, 103, 391-411.

Moskowitz, J. M., Schaps, E., Schaeffer, G. A., &Malvin, J. H. (1984). Evaluation of a substanceabuse prevention program for junior high schoolstudents. International Journal of the Addictions,19, 419^30.

Newcomb, M. D., & Bentler, P. M. (1986). Frequencyand sequence of drug use: A longitudinal study fromearly adolescence to young adulthood. Journal ofDrug Education, 16, 101-120.

O'Connor, J., & Saunders, B. (1992). Drug education:An appraisal of a popular preventive. InternationalJournal of the Addictions, 27, 165-185.

Getting, E. R., & Beauvais, F. (1986). Peer clustertheory: Drugs and the adolescent. Journal ofCounseling and Development, 65, 17-22.

Pentz, M. A. (1985). Social skills and self-efficacy asdeterminants of substance abuse in adolescence. InT. A. Willis & S. Shiftman (Eds.), Coping andsubstance use (pp. 117-142). San Diego, CA:Academic Press.

Rohsenow, D. J. (1983). Drinking habits and expec-tancies about alcohol's effects for self versus others.Journal of Consulting and Clinical Psychology, 51,752-756.

Schaps, E., Moskowitz, J. M., Condon, J., & Malvin,J. H. (1982). Process and outcome evaluation of adrug education course. Journal of Drug Education,12, 353-364.

Schaps, E., Moskowitz, J. M., Malvin, J. H., &Schaeffer, G. A. (1986). Evaluation of sevenschool-based prevention programs: A final report onthe Napa Project. International Journal of theAddictions, 21, 1081-1112.

Scheier, L. M., & Botvin, G. J. (1995). Effects ofearly adolescent drug use on cognitive efficacy inearly-late adolescence: A developmental structuralmodel. Journal of Substance Abuse, 7, 379-404.

Seltzer, V. C. (1989). The psychosocial worlds of theadolescent: Public and private. New York: Wiley.

Sher, K. J., Walitzer, K. S., Wood, P. K., & Brent, E.E. (1991). Characteristics of children of alcoholics:Putative risk factors, substance use and abuse, andpsychopathology. Journal of Abnormal Psychology,100, 427-448.

Smith, G. T., Goldman, M. S., Greenbaum, P. E., &Christiansen, B. A. (1995). Expectancy for socialfacilitation from drinking: The divergent paths ofhigh-expectancy and low-expectancy adolescents.Journal of Abnormal Psychology, 104, 32—40.

Snow, D. L., Tebes, J. K., & Arthur, M. W. (1992).Panel attrition and external validity in adolescentsubstance use research. Journal of Consulting andClinical Psychology, 60, 804-807.

Sobel, M. E. (1987). Direct and indirect effects in

Page 17: Expectancies as Mediators of the Effects of Social ...wordpress.larsri.org/...Botvin-PAB-Expectancies.pdf · expectancies and their encoding into memory are a natural consequence

64 SCHEIER AND BOTVIN

linear structural equation models. SociologicalMethods and Research, 16, 155-175.

Stacy, A. W., Bender, P. M., & Flay, B. R. (1994).Attitudes and health behavior in diverse popula-tions: Drunk driving, alcohol use, binge eating,marijuana use, and cigarette use. Health Psychol-ogy, 13, 73-85.

Stacy, A. W., Leigh, B. C., & Weingardt, K. R. (1994).Memory accessibility and association of alcohol useand its positive outcomes. Experimental andClinical Psychopharmacology, 2, 1-14.

Stacy, A. W., MacKinnon, D. P., & Pentz, M. A.(1993). Generality and specificity in health behav-ior: Application to warning-label and social influ-ence expectancies. Journal of Applied Psychology,78, 611-627.

Stacy, A. W., Newcomb, M. D., & Bentler, P. M.(1991). Cognitive motivation and drug use: A9-yearlongitudinal study. Journal of Abnormal Psychol-ogy, 100, 502-515.

Stacy, A. W., Newcomb, M. D., & Bentler, P. M.

(1992). Interactive and higher-order effects of socialinfluences on drug use. Journal of Health andSocial Behavior, 33, 226-241.

Stacy, A. W., Widaman, K. F., Hays, R., & DiMatteo,M. R. (1985). Validity of self-reports of alcohol andother drug use: A multitrait-multimethod assess-ment. Journal of Personality and Social Psychol-ogy, 49, 219-232.

Stacy, A. W., Widaman, K. F., & Marlatt, G. A.(1990). Expectancy models of alcohol use. Journalof Personality and Social Psychology, 58, 918-928.

Tanaka, J. S. (1987). "How big is big enough?":Sample size and goodness of fit in structuralequation models with latent variables. Child Devel-opment, 58, 134-146.

Webb, J. A., Baer, P. E., Francis, D. J., & Caid, C. D.(1993). Relationship among social and intraper-sonal risk, alcohol expectancies, and alcohol usageamong early adolescents. Addictive Behaviors, 18,127-134.