Attitudes and Health Behavior in Diverse Populations ...people.oregonstate.edu/~flayb/MY PUBLICATIONS/Multiple behaviors... · The affect-based definition of attitude, in which an

Health Psychology1994, Vol. 13, No. 1,73-85

Copyright 1994 by the American Psychological Association, Inc., andthe Division of Health Psychology/0278-6133/94/$3.00

Attitudes and Health Behavior in Diverse Populations: Drunk Driving,Alcohol Use, Binge Eating, Marijuana Use, and Cigarette Use

Alan W. Stacy, Peter M. Rentier, and Brian R. Flay

Five different health behaviors (cigarette use, alcohol use, binge eating, illicit drug use, and drunkdriving) were studied prospectively in 5 different groups of subjects. Associations between attitudestoward these behaviors and the behaviors themselves were investigated over at least 2 waves ofmeasurement. Findings revealed that attitudes predicted behavior nonspuriously in 2 instances:alcohol use and marijuana use. Attitudes did not predict drunk driving, binge eating, or smokingbehaviors. Past behavior predicted attitude in the domains of binge eating and smoking, but not inthe domains of alcohol use, drunk driving, or marijuana use. The results are discussed in terms ofseveral alternative approaches that have implications for interventions that attempt to influencehealth behavior through attitude change.

Key Words: attitudes, health, behavior, longitudinal, expectancy, drugs

Programs designed to promote changes in health behaviorfrequently include components that attempt to change atti-tudes concerning the targeted behavior. Sometimes the focuson attitude change is an explicit component, in which anindividual's evaluations of the behavior are targeted. In othercases, as in many educational media campaigns (e.g., Flay,1987), several key or "salient" beliefs are the focus of thestrategy, which more implicitly attempts attitude change byfirst trying to change beliefs (cf. Ajzen & Fishbein, 1980).Programs that provide information about the harmful conse-quences of smoking, the dangers of drunk driving, or thenecessity of a balanced diet often represent, implicitly orexplicitly, an attempt to change behavior through attitudechange. These programs may focus on harmful outcomes of anegative health behavior, attempting to make personal evalua-tions (attitude) of the behavior more negative, or on beneficialoutcomes of a positive health behavior, attempting to makeevaluations of the healthy behavior more positive. Because ofthe widespread use and potential of attitude concepts inhealth-behavior research and interventions, it is important todocument the predictive impact of this construct and tointegrate theories of health behavior with empirical findings.

The rationale for attitudinal approaches in health-behaviorinterventions has been strengthened by findings in basicresearch documenting the association between attitudes andbehavior (Bentler & Speckart, 1981; Fazio & Williams, 1986;

Alan W. Stacy, Department of Preventive Medicine, Institute forPrevention Research, University of Southern California, Peter M.Bentler, Department of Psychology, University of California, LosAngeles; Brian R. Flay, Prevention Research Center, School of PublicHealth, University of Illinois, Chicago.

Support for this research was provided by Grant DA01070-20 fromthe National Institute on Drug Abuse. The views expressed herein areours and do not necessarily reflect the opinions of the Public HealthService.

Correspondence regarding this article should be addressed to AlanW. Stacy, Institute for Prevention Research, University of SouthernCalifornia, 1000 S. Fremont, #641, Alhambra, California 91803-1358.

Fishbein & Ajzen, 1975; Speckart & Bentler, 1982). However,comprehensive evaluations of the predictive impact of atti-tudes on health behavior have been rare, and the case forattitude as a strong natural precursor to behavior may not be asstrong as some of the basic research would suggest. Our goalwas to investigate the predictive precedence of attitude andhealth behavior over time, so that the usefulness of the attitudeconstruct in health can be more thoroughly examined. First, itis important to summarize briefly the predominant definitionsof attitude and some of the early research on this construct.

Attitude Definitions and Behavioral Prediction

In the theory of reasoned action (Ajzen & Fishbein, 1980;Fishbein & Ajzen, 1975), attitude is thought to represent ageneral evaluation of an object, and affect is considered themost essential aspect of attitude (also see Thurstone, 1931). Inanother dominant view, attitude is considered to be multidimen-sional, composed of correlated but somewhat independentaffective, cognitive, and conative components (Rosenberg &Hovland, 1960). The affect-based definition of attitude, inwhich an overall evaluation of an object or a behavior isemphasized, is probably the more common one in presentresearch in social psychology and health, and this is thedefinition we use in this article. A focus on this definition ofattitude does not imply that cognitive factors are not importantin health behavior, but instead suggests that cognitive variablesare better construed as separate but related constructs (e.g.,Ajzen & Fishbein, 1980; Stacy, Widaman, & Marlatt, 1990).

Attitudinal research has frequently focused on the strengthor nature of attitude-behavior relations, attempting to shedlight on the predictive validity of attitude concepts. Althoughearly research often found poor correlations between attitudeand behavior (e.g., Corey, 1937; LaPiere, 1934; for review, seeWicker, 1971), Fishbein and Ajzen (1975; Ajzen & Fishbein,1980) suggested that relations between attitudes and behaviorwould be stronger when measurements of these variablescorrespond highly in certain properties. Specifically, Ajzen and

73

74 A. STACY, P. BENTLER, AND B. FLAY

Fishbein's (1977) review of the literature identified strongattitude-behavior associations across a number of studies inwhich at least the target (object) and action (e.g., drinking)elements of measurement corresponded highly.

Fishbein and Ajzen's (1975) approach is consistent with theposition that attitudes have predictive priority over behavior(e.g., Bentler & Speckart, 1981). Other positions relevant tohealth behavior include the view that behavior causes attitude(Bern, 1978), attitudes and behavior influence each otherreciprocally (Kelman, 1974), and the influence of attitude onbehavior depends strongly on moderator variables, such asindividual differences, in the accessibility of attitudes or beliefsfrom memory (Fazio & Williams, 1986; Stacy, Dent, et al.,1990) or in personality characteristics (e.g., Snyder, 1974).Although these latter perspectives have made contributions tothe study of moderators of attitude-behavior relations, mosthealth-behavior campaigns focus on attitude and behaviorchange rather than on attitude-moderator change. Thus, atleast in health-behavior research, it is important to determinethe precedence of attitudes and behavior as direct-effectpredictors of one another. Surprisingly, questions about thepredictive priority of attitudes and behavior have been ad-dressed rarely in health-behavior research.

Attitudes and Specific Health Behaviors

Smoking

The influence of attitudes on smoking onset and cessation isimportant because attitude-change strategies are frequentlyused in large mass media programs (Farquhar, 1991; Flay,1987) and school-based interventions (e.g., Bruvold, 1990).Although little is known at present about the critical elementsof successful interventions in this area (Flay, 1987; Snow,Gilchrist, & Schinke, 1985), theoretically sound attitude-change strategies are in principle quite amenable to massmedia efforts and school-based interventions. Because mostadult smokers begin to smoke in adolescence (Flay, d'Avernas,Best, Kersell, & Ryan, 1983; Leventhal & Cleary, 1980), thisage group seems particularly important for research andintervention. Charlton and Blair (1989; but see Jarvis, God-dard, & McNeil, 1990) reported evidence that attitudes amonginitial nonsmokers predicted adolescent smoking behaviorprospectively, but they did not investigate the predictiveprecedence of attitude and behavior over time. In otherinstances, attitudes that do not correspond in level of specific-ity to this behavioral target and action (smoking) have beenstudied in smoking research, including attitudes toward self ortoward health (e.g., Brunswick & Messeri, 1984). Consistentwith Ajzen and Fishbein's (1977) predictions and findings, alack of correspondence in the target or action element ofattitudes and behavior has led to low attitude-behavior relation-ships. Despite the widespread use of attitude concepts in bothinterventions and theories of adolescent smoking, no studieshave investigated the predictive precedence of attitudes to-ward smoking and smoking behavior in this population.

Alcohol Use and Drunk Driving

Existing evidence about attitude-behavior relations in alco-hol-use research can be considered mixed. Kahle and Herman(1979) found that attitudes showed predictive predominanceover behavior in a cross-lagged panel correlational investiga-tion of alcohol consumption over a 2-month period (also seeSchlegel, Manske, & d'Avernas, 1985), but this type of statisti-cal analysis has been criticized on several grounds (Rogosa,1980). In a structural equation analysis, Bentler and Speckart(1979; also see Stacy, Widaman, & Marlatt, 1990) foundevidence that attitude toward alcohol use predicted drinkingbehavior 2 weeks later, although they did not perform anexamination of predictive precedence. In a 3-year prospectivestudy, Johnson (1988) found no evidence that attitudes aboutdrinking and drinking behavior predicted one another overtime, in either of the possible directions. No studies haveinvestigated the predictive precedence of attitudes and behav-ior regarding alcohol-related problem behaviors, such asdriving under the influence of alcohol (DUI). However, thecall for attitude change regarding alcohol use frequentlyaccompanies discussions of the social costs of this and otherproblems associated with alcohol use.

Illicit Drug Use

As with tobacco and alcohol use, many intervention effortsattempt some form of direct or indirect attitude change in theprevention and treatment of drug abuse (e.g., Pentz et al.,1989). Although attitudes appear to be predictive of certaintypes of illicit drug use (e.g., Akers, Krohn, Lanza-Kaduce, &Radosevich, 1979), few studies have examined whether thisprediction is as strong as the prediction of attitudes by druguse. If the prediction of attitudes by drug use is much strongerthan the prediction of drug use by attitudes, then it may be thecase that behavior must change before attitude changes. Apredictive predominance of behavior over attitude would haveimplications for the focus of interventions.

The only existing evidence concerning the predictive prece-dence of attitudes and illicit drug use was reported by Johnson(1988) and by Schlegel et al. (1985). Johnson found thatattitude and marijuana use did not predict one another overtime, and that only the stability (autoregressive) effects of thesame construct on itself over time were significant. Schlegel etal. reported only preliminary information about predictiveprecedence, as evaluated in a cross-lagged correlational designof marijuana attitudes and behavior.

Binge Eating

During the past two decades, binge eating has appeared toincrease in frequency among young women (Polivy & Herman,1985). Although the behavior of binge eating may be part ofthe larger eating disorder of bulimia, people who binge eatmay vary continuously in their extent of disordered eatingbehavior and concomitant symptoms. Research investigatingthis behavior is important because of a number of negativehealth consequences of the behavior (e.g., Hawkins & Clem-ent, 1980), whether or not a clear, clinically relevant case of

ATTITUDES AND HEALTH BEHAVIOR 75

bulimia is evident. Even when binge eating is not apparentlysevere, the typical diet associated with binge eating is high infat and is likely to cause health consequences if the dietcontinues for an extended period of time.

A number of studies have found that cognitive factors moreconsistently trigger binge-eating behaviors than do physiologi-cal variables, such as caloric intake (e.g., Polivy, Herman,Olmstead, & Jazwinski, 1984; Spencer & Fremouw, 1979), andthat certain types of attitudes are associated with this behavior.However, the only types of attitudes that have been investi-gated are attitudes about one's body (for review, see Ben-Tovim & Walker, 1991). Attitudes toward the target behavior(e.g., Fishbein & Ajzen, 1975) have not been investigated as apredictor of binge eating, so the degree to which binge eating isunder attitudinal control is unknown at present. Thus, al-though attitudinal approaches to primary prevention of bingeeating among adolescents and young adults may be theoreti-cally feasible, the absence of empirical evidence showing arelation between attitudes and behavior in this domain damp-ens the potential enthusiasm for such an approach.

Summary

Little is known about the precise relations between attitudeand the health behaviors we address in this article. Whenattitudes have been studied, attitude and behavior usually havenot been measured prospectively over time in a way that wouldreveal predictive precedence of one variable over the other(e.g., Bentler & Speckart, 1981). We measured binge eating,alcohol use, drunk driving, cigarette smoking, and illicit druguse behaviors and attitudes in five different groups sampledfrom populations in which these problem health behaviors arerelevant. Although there are many other variables relevant tothe study of attitude-behavior relations (e.g., Fazio & Wil-liams, 1986; Madden, Ellen, & Ajzen, 1992; Stacy, Widaman,& Marlatt, 1990), the only relevant variables available acrosseach of the present studies were measures of attitude andbehavior. As stated earlier, the investigation of the predictiverelations between attitude and health behavior is important inits own right, especially when a range of different healthbehaviors are investigated prospectively. Furthermore, manyhealth-promotion interventions using attitude-change strate-gies have assumed implicitly that attitudes have direct effectson behavior, consistent with the predictive models we evalu-ated.

General Method

In each study reported below, we measured attitude and behaviorprospectively and used covariance structure analysis (e.g., Bentler,1989; Bentler & Speckart, 1981; Bollen, 1989) to investigate thepredictive precedence of the two primary variables. We report resultsbased on maximum likelihood estimation. This form of estimation hasbeen found to be relatively robust to departures in multivariatenormality (e.g., Huba & Harlow, 1987), but we also evaluated each ofthe covariance structure models using at least one other method ofestimation. We evaluated the statistical significance of paths usingchi-square difference tests and critical ratios; the overall goodness offit of the models was evaluated primarily through use of the nonnor-

med fit index (NNFI; Bentler & Bonett, 1980), which has been foundto be among the most robust indexes of fit regarding sample-size biases(Marsh, Balla, & McDonald, 1988). We also report the normed fitindex (NFI; Bentler & Bonett, 1980).

In each of the studies, measures of attitude and behavior corre-sponded specifically in at least the target (e.g., cigarettes) and action(e.g., smoking) elements of measurement, which have been found tobe most important in the study of attitude-behavior relations (Ajzen &Fishbein, 1977). The context element (e.g., place where smokingoccurs) remained quite general in all of the studies, but correspondedacross measures of attitude and behavior. The time element (e.g.,smoking in next month) corresponded very specifically in the threeshort-term studies and more generally in the two longer-term studies.

For each behavioral domain we studied, only subjects who indicatedthey had engaged in the relevant behavior were used in the analysis.Thus, the different samples were equated for having at least minimalexperience with the behavior. Equating the samples in this wayreduced the possibility that attitude-behavior relations differed acrosssamples because of differences in relevant experience—a likely mod-erator of attitude effects. A number of findings in basic research showthat this moderator confound is likely, because subjects who differ indirect experience with an object typically show different strengths ofattitude-behavior relations (for review, see Raden, 1985). In thesestudies, the subjects with direct experience generally showed strongerlevels of prediction of behavior by attitudes. Clearly, some type ofcontrol of level of experience is necessary if results from differentsamples with varying levels of experience are to be compared. Thesubjects in each study were predominantly middle class and White, butrepresented several different age and nationality groups.

Study 1: Binge Eating

Method

Subjects. Respondents were 260 undergraduate female collegestudents who participated in the study for extra credit in introductorypsychology classes. Although 241 subjects completed all three waves ofmeasurement and provided complete data, only the subjects whoparticipated at all waves and who indicated they binge ate at some timein the past year (N = 144) were included in data analysis. We sampledonly women because this problem health behavior generally is muchmore frequent in women than in men.

Procedure. The survey was given at three different times, separatedby 2-week intervals. Respondents completed questionnaires at predes-ignated times in a classroom, group-testing situation. Questionnaireidentification numbers were used to allow for matching of question-naires across waves; these numbers were self-generated by the subjectsaccording to a previously pilot-tested protocol that ensures a high levelof matches (95% or more) over a short period. The anonymity of thequestionnaire was emphasized both by the administrator and inwritten statements in the questionnaire. Previous research has shownthat if anonymity is guaranteed fully, valid responses to sociallyproscribed behaviors are likely even among young adolescents provid-ing self-reports of drug use (Murray & Perry, 1987). Strong assurancesof anonymity appear to be as effective as are bogus pipeline proce-dures in studies of consummatory behavior (Murray & Perry, 1987).

Measures of behavior. The behavioral measures at each measure-ment asked respondents to report on their binge eating in the past 2weeks. Before answering the binge-eating questions, respondents wereasked to read a common definition of binge eating: "Binge eating is therapid consumption of a large amount of food in a discrete period oftime, usually less than two hours." For this behavior, 18 activities were


assessed that characterize the behavior, consistent with criteria fromthe Diagnostic and Statistical Manual of Mental Disorders (3rd ed., rev.;Z>SM-///-rR; American Psychiatric Association, 1987) and frequentlyused definitions (Katzman & Wolchik, 1984; Polivy et al., 1984). Theseitems were consistent with most criteria of binge eating based on thesesources, but DSM-III-R criteria that refer more specifically to bulimiawere omitted (e.g., dieting and purging). Thus, in our study, bingeeating was used as a more general construct than bulimia per se. Mostitems were scored from 1 to 7, with frequency items anchored by not atall (1) and every day (7). The binge items assessed the number of daysrespondents: binge ate, ate large amounts of food while feelingcompelled to do so, were out of control in their binge eating, hid bingeeating from others, could not stop eating voluntarily, ate an extremeamount of a single type of food, consumed more than 1,200 caloriesper binge, ate much more than needed beyond satisfying hunger, atemuch more than typical for body size, ate high-caloric and easilyingestible food, felt depressed after binge eating, kept eating until theythought they would explode, ate to the point of feeling sick, ate a largeamount of food when not hungry, felt food controlled their life, and atemuch more rapidly than most people ever eat. An additional itemasked respondents how much food they ate very rapidly when theybinge ate, coded from 1 (none) to 7 (a very large amount). A self-ratingitem (see Stacy, Widaman, Hays, & DiMatteo, 1985) assessed howrespondents rated themselves regarding their binge eating, from 1(non-binge eater) to 8 (very heavy binge eater).

To organize the binge items into factors, we composed threeindicators of binge eating by randomly assigning each of the aforemen-tioned binge items to one of the indicators, resulting in three indicatorscomposed of simple sums of 6 items each. We conducted a preliminarystudy to evaluate the construct validity of these measures. In thatstudy, the binge-eating measures were found to show adequate levelsof item convergence and discrimination when thorough multitrait-multimethod procedures were used (Stacy & Saetermoe, 1987). Forexample, binge-eating indicators did not load on other behavioralfactors (dieting and alcohol use), all binge-eating loadings weresubstantial (above .81), and none of the potential assessment methodfactors appeared to bias the results.

Measures of attitude. Attitude at each time of measurement wasassessed with three indicators that each were simple sums of two items.The items composing this scale were introduced with the statement"Your binge eating in the next four weeks is": followed by six 7-pointbipolar adjective items ranging from extremely good to extremely bad,extremely beneficial to extremely harmful, extremely rewarding to extremelypunishing, extremely pleasant to extremely unpleasant, extremely enjoyableto extremely unenjoyable, and extremely favorable to extremely unfavor-able. These measures are based on the recommendations of Ajzen andFishbein (1980).

Cross-lagged structural models. All of the models we tested used thesame measurement model, in which the scale of measurement wasdetermined by fixing Tl factor variances at 1.0 and imposing equalityconstraints on loadings of identical indicators of the same constructmeasured at the three different times. In addition, the measurementmodel specified covariances among unique scores of identical indica-tors measured across the three times. These procedures are consistentwith those of Joreskog (1979).

Results

The initial, structurally saturated model specified all fourpossible predictive paths from Tl attitude and binge factors toT2 attitude and binge factors. An analogous set of four pathswas estimated among T2 and T3 attitude and binge factors andamong Tl and T3 attitude and binge factors. Covariances

between same-wave factors at Tl and same-wave residualfactor covariances at T2 and T3 also were freely estimated. Weused the chi-square value for this initial model (x2 [110,N = 144] = 144.98,;? = .01) in model comparisons with more-restricted models. Thus, the paths in the initial model weretested for statistical significance by deleting paths one at a timefrom the initial model and evaluating the decrease in fit fromthe initial model in single-degree-of-freedom chi-square differ-ence tests (cf. Bentler & Speckart, 1981). Using this proce-dure, all nonsignificant (p < .05) paths, factor covariances,and factor residual covariances were deleted from the initialmodel, deriving the final model. This final model (x2 [118,N = 144] = 154.97,;? = .01) fit the data well regarding practi-cal indices of fit (NNFI = .99; NFI = .96). An alternativeestimator appropriate for relatively small samples, using ellip-tical distribution theory (Shapiro & Browne, 1987), yieldedidentical results regarding significance tests of alternativepaths.

No single path, factor covariance, or factor residual covari-ance could be added to this final model to increase its fitsignificantly, nor could any single estimate in the final model beomitted without resulting in a significant decrease in model fit.In a more global test, the final model was not significantlydifferent in fit from the initial model (x2 [8, N = 144] = 9.98,p > .05). The only significant cross-lagged path in the finalmodel was from Tl behavior to T2 attitude. In addition,covariances between Tl attitude and behavior factors andbetween T3 attitude and behavior factor residuals were nonsig-nificant. The final model is shown in Figure 1, which presentsthe standardized structural parameter estimates and residualvariances. Standardized factor loadings in the measurementmodel all were statistically significant (allps < .001) and wereof adequate size, ranging from .71 to .96 for attitude and from.96 to .98 for binge-eating. These moderate to large loadingsindicated that these indicators were measured with goodinternal consistency; they were well above the minimumloading sizes recommended prior to interpretation in factoranalysis (e.g., Gorsuch, 1983).

Study 2: College Alcohol Use

Method

Subjects. Respondents were 356 male and female undergraduatecollege students enrolled at a small West Coast university. Of the total,334 respondents provided complete data and completed both waves ofmeasurement. A subsample of respondents who indicated that theyhad drunk at least some alcohol in the previous year was selected foranalysis (N = 270). The students received extra credit in an introduc-tory-level psychology course for their participation in the study. Thesedata were part of a larger study on alcohol use (Stacy, Widaman, &Marlatt, 1990). In the previous study, we did not examine a model inwhich behavior could predict attitude over time. Because the presentcross-lagged structural model, which is parallel to the other modelsanalyzed in this article, was not analyzed in the previous study, it isappropriate to report these analyses here.

Procedure. The procedure was identical to the one reported inStudy 1, except that we assessed each subject at two different times,separated by a 4-week interval. Anonymity of responses was again

ATTITUDES AND HEALTH BEHAVIOR

. 2 8 * * *

77

BingeEatingTime 3

Binge»>( Eating

Time 2

BingeEatingTime 1

. 7 6 * * * / Att i tudeTime 3

Att i tudeAtt i tudeTime 1

Figure 1. Final structural equation model and standardized maximum likelihood estimates for bingeeating. (Large circles designate latent constructs. Small circles depict residual variances of factors. Pathcoefficients are depicted with single-headed arrows. Correlations between factors or factor residuals aredepicted with double-headed arrows. Significance levels are based on critical ratios on unstandardizedmaximum likelihood estimates [*p < .05; **p < .01; ***p < .001].)

thoroughly emphasized. Three indicators were developed for eachlatent variable.

Measures. Measures of attitude were identical to those outlined inStudy 1, but were worded in terms of alcohol use. That is, subjectsresponded to identical semantic differential attitude items but re-ported attitudes in terms of "your drinking alcohol in the next fourweeks." Alcohol-use measures were constructed to assess the generalextent (i.e., both quantity and frequency) of alcohol use during the past4 weeks. The first indicator of alcohol use was the sum of two 7-pointscales, closely modeled after scales used by Bentler and Speckart(1979,1981). Both scales assessed frequency of use; one had endpointsof every day and not at all, and the second scale had endpoints ofextremely frequently and never. The second indicator used an 8-pointrating scale on which each respondent was instructed to classifyhimself or herself into 1 of 8 drinker categories (very heavy drinker,heavy drinker, fairly heavy drinker, moderate drinker, fairly lightdrinker, light drinker, very light drinker, and nondrinker). The thirdindicator was a quantity-frequency, or QF, index in which quantity andfrequency of typical use (number of drinks consumed in previousmonth, based on typical quantity consumed on days the subject drank)was summed with greatest use (greatest number of drinks the subjectconsumed multiplied by the number of days the subject drank thisgreatest amount). A thorough multitrait-multimethod assessment ofsimilar alcohol indicators has demonstrated their convergent anddiscriminant validity (Stacy et ah, 1985). In addition, Stacy et al. foundevidence that method effects, such as general self-report biases, werenot strong influences on this type of data.

Cross-lagged structural models. The model construction and evalua-tion in this study were identical to that outlined in Study 1, except thatthis model involved a two-wave design.

Results

The initial, structurally saturated model specified all fourpossible predictive paths from Tl attitude and alcohol factorsto T2 attitude and alcohol factors. Covariances betweensame-wave factors at Tl and same-wave residual factor covari-ances at T2 also were freely estimated. We used the chi-squarevalue for this initial model (x2 [46, N = 270] = 73.40, p < .01)in model comparisons with more-restricted models, using thesame model testing and parameter trimming procedures asreported in Study 1. Only one path, from Tl behavior to T2attitude, was nonsignificant using this procedure, and this pathwas eliminated to derive the final model. The final model (x2

[47, N = 270] = 74.98, p < .01) fit the data well according topractical indices of fit (NNFI = .98; NFI = .97). Ellipticaltheory and arbitrary distribution theory estimation (Bentler,1989) led to the same final model. The final model did notdiffer significantly from the initial model. The standardizedstructural parameter estimates for the final model are pro-vided in Figure 2. Standardized factor loadings ranged from


. 25 '

Figure 2. Final structural equation model and standardized maxi-mum likelihood estimates for college alcohol use. (Large circlesdesignate latent constructs. Small circles depict residual variances offactors. Path coefficients are depicted with single-headed arrows.Correlations between factors or factor residuals are depicted withdouble-headed arrows. Significance levels are based on critical ratioson unstandardized maximum likelihood estimates [*p < .05;**/> < .01;"**p < .001].)

.82 to .86 for attitude and from .8ps < .001).

I to .93 for alcohol use (all

Study 3: Alcohol Use and Drunk DrivingAmong DUI Offenders

Method

Subjects. Respondents were 175 men who had been convicted ofdriving under the influence (DUI) of alcohol and who were participat-ing in an educational program mandated by the state of California.Because all of the subjects drank alcohol and had engaged in DUI inthe recent past, no subjects were removed from the analysis. Out ofthis sample, 164 respondents completed both waves of measurement.Respondents were paid $5.00 to complete the questionnaire, whichwas completely voluntary and anonymous.

Procedure. The procedure was identical to the one described inStudy 2. Subjects again were assessed at two different times, separatedby a 4-week interval. No subjects completed the program before thefinal assessment. Respondents completed questionnaires on a volun-tary basis after regular driver-education program meetings, in aclassroom setting. Anonymity of responses was again thoroughlyemphasized, but respondents were further assured that their participa-tion in the study had nothing to do with the education program andthat no one, including the program staff, had any way to discover theirresponses.

Measures of alcohol use behavior and attitude. Three indicatorswere developed for each latent variable. Measures of attitude towardalcohol use were identical to those outlined in Study 2, except that oneof three attitude indicators contained only one item because only fiveattitude items were used in the questionnaire. Alcohol-use measuresincluded the three indicators used in Study 2, using the samefrequency, rating, and QF methods of measurement; similar alcohol-use indicators were validated by Stacy et at. (1985).

Measures of DUI behavior and DUI attitude. Attitude towarddriving under the influence of alcohol was assessed with a singlebipolar item that asked respondents to answer the following questionon a scale coded from 1 (extremely unfavorable) to 7 (extremely

favorable): "Your attitude toward or feeling about your driving afterdrinking in the next month is?" This item is in the form of one of thealternative methods of measuring attitude suggested by Ajzen andFishbein (1980). DUI behavior was measured by three items. One itemasked respondents how often they drove after drinking alcohol in thepast month, coded from 1 (not at all) to 7 (every day). A second itemasked respondents how often they drank at least three drinks rightbefore driving in the past month, coded on the same 7-point scale. Thethird item asked respondents to rate what kind of drinker they werewhen they drove after drinking in the last month, coded from 1(nondrinker) to 8 (very heavy drinker).

Cross-lagged structural models. Two different domains of attitude-behavior relations (DUI and alcohol use per se) were examined in twodifferent series of models. For the alcohol-use domain, the models wetested used the same model construction and evaluation procedures asin Studies 1 and 2. For the DUI domain, the behavior factor at Tl andT2 represented three indicators, as with the aforementioned alcoholfactor, but the attitude factors at both times were composed of a singleindicator, consistent with a standard path-analytic specification (e.g.,Bentler, 1989). The model-evaluation procedure for the DUI domainwas the same as the procedure used in the alcohol-use domain and inStudies 1 and 2.

Results

Alcohol Use. The chi-square values for the initial, satu-rated structural model (x2 [46, N = 164] = 53.44, p = .21)were employed in model comparisons with more-restrictedmodels, using the same model-testing and parameter-trimmingprocedures as reported in the previous studies. The twocross-lagged paths and the covariance between the factorresiduals at T2 were found to be nonsignificant. These threeestimates were omitted to derive the final model. The finalmodel (x2 [49, N = 164] = 57.68,p = .19) fit the data very wellusing practical indices of fit (NNFI = .99; NFI = .97). Thefinal model did not differ significantly from the initial model interms of model fit, and the significance of paths did not changewhen we evaluated the models with alternative (elliptical andarbitrary distribution) estimators. Standardized structural pa-rameter estimates for the final model are reported in Figure 3.Standardized factor loadings ranged from .81 to .97 forattitude and from .89 to .96 for alcohol use (all/?s < .001).

Driving under the influence of alcohol. The initial model wasstructurally saturated, as in the initial model for alcohol use.The chi-square value for this initial model (x2 [15,N = 164] = 50.00, p < .001) was to be used in model compari-sons with more-restricted models, but more-restricted modelscould not be estimated, because of linear dependencies inestimation. Even when a model is mathematically identified,such dependencies can be a product of "empiricalunderidentification" (resulting from various peculiarities inthe data; Rindskopf, 1984). However, the structurally satu-rated model did not contain any dependencies or estimationproblems, and the statistical significance of parameter esti-mates therefore could be evaluated on the basis of z-tests fromthis model (instead of model-trimming procedures). Neither ofthe cross-lagged paths, from attitude to behavior or frombehavior to attitude, was significant at p < .05 according tothese tests. Standardized structural parameter estimates fromthis model are provided in Figure 4. Standardized factorloadings on DUI behavior ranged from .70 to .95 and each was


statistically significant (all ps < .001); because it was a singleindicator, DUI attitude had no factor loadings. This model fitthe data somewhat less well than did the preceding finalmodels (NNFI = .92; NFI = .94). Use of the alternative esti-mators did not change the pattern of significance of structuralpaths.

Study 4: Adolescent Smoking

Method

Subjects. Respondents were 199 male and female llth-gradeCanadian high-school students who indicated they had smoked at leastonce in their lives. These students constituted a subsample of a largersample of students involved in a longitudinal study of adolescentsmoking described by Flay et al. (1985).

Procedure. We assessed each subject at two different times, sepa-rated by approximately 12 months. Respondents completed question-naires in regular high school classrooms. Confidentiality, but notanonymity, of responses was emphasized. A number of studies haveindicated that self-reports of cigarette use among adolescents arereasonably valid as long as certain safeguards are present duringquestionnaire administration (e.g., Bauman & Dent, 1982; Evans,Hansen, & Mittelmark, 1977; Murray & Perry, 1987; Pechacek et al.,1984). For example, Murray and Perry recommended that pipelinemethods (cf. Jones & Sigall, 1971), which involve obtaining biologicalspecimens, should be used to increase the validity of self-reports inadolescents when anonymity cannot be guaranteed. Thus, we used apipeline procedure in which carbon monoxide (CO) measures weretaken from each student to increase the validity of self-reports ofcigarette use. Stacy, Flay, et al. (1990) found that this procedureyielded an adequate level of convergent validity between self-reportedcigarette use and CO measurement in adolescents from the samepopulation of subjects as sampled in this study.

DUI-OffenderDrinkingTime 1

DUI-OTfenderDrinkingTime 2

.70 '

Figure 3. Final structural equation model and standardized maxi-mum likelihood estimates for driving under the influence (DTJI)-offender alcohol use. (Large circles designate latent constructs. Smallcircles depict residual variances of factors. Path coefficients aredepicted with single-headed arrows. Correlations between factors orfactor residuals are depicted with double-headed arrows. Significancelevels are based on critical ratios on unstandardized maximumlikelihood estimates ['p < .05; **p < .01; ***p < .001].)

DUI-OffenderDrinking

DrivingTime 1

. 6 0 *

AttitudeTime 1

.33*" AttitudeTime 2

DUI-OffenderDrinking-DrlvlngTime 2

.21'

Figure 4. Final structural equation model and standardized maxi-mum likelihood estimates for DUI-offender drinking-driving. (Largecircles designate latent constructs. Rectangles represent measured-variable, or single-indicator, constructs. Small circles depict residualvariances of constructs. Path coefficients are depicted with single-headed arrows. Correlations between constructs or their residuals areshown with double-headed arrows. Significance levels are based oncritical ratios on unstandardized maximum likelihood estimates["p < .05; "p < .01; ***p < .001].)

Measures of attitude. Measures of attitude were quite similar to theones used in the previous studies. Attitude was assessed with threeindicators that were each simple sums of two items. The itemscomposing this scale were introduced with the statement "Smokingis": followed by six 5-point bipolar adjective items ranging from verybeautiful to very ugly, very good to very bad, very clean to very dirty, verysafe to very dangerous, very pleasant to very unpleasant, and very nice tovery awful.

Measures of smoking. Three different indicators of smoking wereused. The first indicator was the smoking index used by Biglan,Gallison, Ary, and Thompson (1985), which is computed on the basisof the average of two items, including open-ended scales assessing last24-hour use and past-7-day use. The other two indicators wereproposed by the National Cancer Institute (NCI, 1986). One of theseindicators was a closed-ended, 10-point scale of current smoking thatcombines frequency and quantity, asking respondents "How much doyou currently smoke?" (I have never smoked = 1, Not at all in the last12 months = 2, A few times in the last 12 months = 3, Usually once amonth = 4, A few times each month = 5, Usually once a week = 6, Afew times each week = 7, A few times most days = 8, About half apack each day = 9, A pack or more each day = 10.). The second NCIindicator of smoking was a closed-ended item with 7 points, askingrespondents "How frequently have you smoked cigarettes during thepast 30 days?" (Not at all = 1, Less than one cigarette per day = 2,One to five cigarettes per day = 3, About one-half pack per day = 4,About one pack per day = 5, About one and a half packs per day = 6,two packs or more per day = 7.). Previous research has demonstratedthe convergent and discriminant validity of these types of measureswhen used as continuous indexes (Stacy et al., 1985; Stacy, Flay, et al.,1990).

Cross-lagged structural models. The model construction and evalua-tion procedures of Studies 1 and 2 were used in this study for the initialand subsequent models.

Results

We used the initial, structurally saturated model (x2 [46,N = 199] = 132.42,p < .001) in model comparisons with more-


restricted models, using the procedures reported in the previ-ous studies. One cross-lagged path, from Tl attitude to T2smoking, was found to be nonsignificant; this path was omittedto derive the final model. The final model (x2 [47,N = 199] = 132.50, p < .001) fit the data well according topractical indices of fit (NNFI = .96; NFI = .95). The finalmodel did not differ significantly from the initial model interms of model fit. Standardized structural parameter esti-mates for the final model are reported in Figure 5. Standard-ized factor loadings ranged from .84 to .91 for attitude andfrom .90 to .97 for smoking (all/js < .001). Model evaluationsusing alternative (elliptical and arbitrary distribution) estima-tors showed the same pattern of significance of structuralpaths.

Study 5: Illicit Drug Use

Method

Subjects. Respondents were 706 male and female young adultswho were participants in a larger study on drug abuse etiology andconsequences. The predominantly White, middle-class sample had anaverage age of 17.9 years at the first time of measurement used in thisstudy, which was the third wave of assessment in the larger study(Newcomb & Bentler, 1988). The subjects were recruited for participa-

tion in the larger study from representative junior high schools fromLos Angeles County.

Procedure. We assessed each subject at two different times, sepa-rated by approximately five years. Respondents completed a self-administered questionnaire at both times of measurement. Althoughthe data were not anonymous, the complete confidentiality of re-sponses was emphasized to respondents, who were informed that theirresponses were protected by a federal certificate of confidentiality.

Measures of attitude and behavior. The two measures of attitudeasked respondents to indicate what they thought about smokingmarijuana or taking PCP, on scales from 1 (terrible idea) to 5 (greatidea). These measures were found to show an adequate degree ofreliability in previous research as shown in high loadings on theirrespective factors (Bentler & Speckart, 1979). The measures of illicitdrug-use behavior assessed the frequency of use of a variety ofdifferent drugs over the past 6 months on scales from 1 (never) to 7(more than once a day). Although we assessed use of a variety ofdifferent drugs, we restricted the primary analyses to marijuana usebecause of results from some preliminary analyses described below.

Cross-lagged structural models. The model-evaluation procedureswere the same as those used in the previous studies. However, theconstruction of the primary model was different because of someresults from a preliminary analysis. As outlined above, attitudemeasures were available only for marijuana and PCP use. In thepreliminary analysis, we attempted to model attitude-behavior rela-tions between a general factor of drug-use attitudes, representing the

AdolescentSmokingTime 1

AdolescentSmokingTime 2

. 4 0 * * *

Figure 5. Final structural equation model and standardized maximum likelihood estimates for adoles-cent smoking. (Large circles designate latent constructs. Small circles depict residual variances of factors.Path coefficients are depicted with single-headed arrows. Correlations between factors or factor residualsare depicted with double-headed arrows. Significance levels are based on critical ratios on unstandardizedmaximum likelihood estimates [*p < .05; **/> < .01; ***p < .001].)


common variance of marijuana and PCP attitudes, and more generaldrag-use behavior (a variety of different drugs). However, on the PCPmeasures only a few respondents indicated values above the minimum.In addition, the preliminary model did not yield any significant pathsbetween the general factors, although it did support the significance ofsome paths between the marijuana indicators of these factors. Becauseonly marijuana indicators had significant effects, and because themeasurement of PCP was problematic, we only report the analysis of amodel that was restricted to the associations between marijuanabehavior and attitude indicators over time. In this primary model,attitude toward marijuana use is represented by a single indicator,rather than as a factor, and marijuana behavior also is represented by asingle indicator. Otherwise, the model was constructed similarly to themodels in the previous studies. To maintain compatibility with theother studies, only subjects who had smoked marijuana at least once atthe initial time of measurement were used in the analysis (N = 357).

Results

The primary path model just described for marijuana useestimated paths from Tl attitude and behavior indicators to T2attitude and behavior indicators, as well as a covariancebetween the Tl indicators and a covariance between the T2residuals of the predicted indicators. Because this initial modelwas completely saturated (i.e., no additional parameter esti-mates could be made in the model), it fit the data perfectly (x2

[0, N = 357] = 0). This model was used in model comparisonswith more-restricted models, using the procedures reported inthe previous studies. One cross-lagged path, from Tl mari-juana use to T2 attitude toward marijuana, was found to benonsignificant; this path was omitted to derive the final model.

The final model (x2 [1, N = 357] = .66, p = .42) fit the data

well according to practical indices of fit (NNFI = 1.00;NFI = 1.00). The final model did not differ significantly fromthe initial model in terms of model fit. Standardized structuralparameter estimates for the final model are reported in Figure6. Because the model used only single indicators rather thanfactors, there were no factor loadings. Model evaluations usingalternative (elliptical and arbitrary distribution) estimatorsshowed the same pattern of significance of structural paths. Inseveral supplementary analyses, adjustments of the assumedamount of error variance in the single-indicator constructs didnot change these results.

Discussion

Many intervention programs include attitude change as animportant element of health-behavior change. In addition,previous research, as well as most of the present studies, hasshown significant cross-sectional correlations between atti-tudes and behavior, suggesting that attitudes may be importantprecursors of health-related behavior. However, the mostgeneral conclusion from the present longitudinal results isthat, when stabilities and cross-sectional correlations weretaken into account, attitude was neither a consistent nor strongpredictor of the health behaviors we assessed. Althoughattitude did not have general or strong effects, it did predictcertain specific health behaviors in two of the six models weanalyzed, suggesting that attitude still may be of some impor-tance.

AdolescentMarijuana Use

Time 1

. 5 5 * * *

AttitudeTime 1

AdultMarijuana Use

Time 2

AttitudeTime 2

60

Figure 6. Final structural equation model and standardized maximum likelihood estimates for marijuanause. (Rectangles represent measured-variable, or single-indicator, constructs. Small circles depict residualvariances of constructs. Path coefficients are depicted with single-headed arrows. Correlations betweenconstructs or construct residuals are depicted with double-headed arrows. Significance levels are based oncritical ratios on unstandardized maximum likelihood estimates [*p < .05;**p < .01; ***p < .001].)


The failure of attitude to be consistently predictive amongsubjects who have a minimal degree of behavioral experiencedoes not imply that affective or cognitive approaches are notimportant in health behavior. On the contrary, the lack ofdirect effects of attitude in most of our analyses suggests thatother approaches involving affective or cognitive constructsshould be considered as viable alternatives to the direct-effectattitude model we investigated. We now outline several of thepossible reasons why attitude may not be a consistent, direct-effect predictor of health behavior.

First, it is possible that attitude toward health behavior isuseful empirically, theoretically, and practically only in thecontext of a more general theory than that implied by ourinvestigation of direct-effect paths from attitude to behavior.More general theories, such as reasoned action (Fishbein &Ajzen, 1975), planned behavior (Ajzen & Madden, 1986;Madden et al, 1992), attitude accessibility (Fazio & Williams,1986), or social-cognitive approaches to attitude (e.g., Wyer &Srull, 1989) contain more complex predictions than the simple,direct-effect paths we were able to investigate. Future researchon attitudes toward health behavior may find attitude moregenerally explanatory in the context of one of these formaltheories. If so, health-behavior interventions may be improvedby an acknowledgment of the complexity of attitude effects onbehavior as well as by the explicit use of propositions fromthese theories in programs of change. Although some health-behavior interventions use empirically supported, theoreticallydriven programs of attitude-behavior change, many others donot go beyond a simple direct-effect model of attitude effects.

There are other available interpretations. Our study usedone of the most typical definitions of attitude, which wasrepresented as a single bipolar affective predisposition towardperforming a behavior. In this approach, a person cannot holdstrong positive and negative attitudes simultaneously withoutresponding to the attitude measure in the middle range of thesingle attitude continuum. This approach assumes that positiveand negative affect toward a behavior converge or balance outin a type of summary of affective tendencies, which theninfluences behavior (or a mediator of behavior, e.g., intention;Fishbein & Ajzen, 1975). One example of such an approachregarding health behavior has been outlined by Akers (1992),in which the stronger affective tendencies win out in the overallattitudinal predisposition toward a behavior. Obviously, peoplecan answer attitude questions in this way, because the bipolarconstruction of attitude scales is likely to force some consider-ation of both positive and negative affect when answering thequestions. However, this does not imply necessarily thatattitude is a unitary entity that is represented in this way in themind. In some perspectives, affect is not a unitary summary ofpositive and negative qualities of a behavior or object, but isinstead two separate dimensions of positive and negativeaffect. These perspectives are in agreement with findingsshowing that affect often appears as two dimensions of positiveand negative (e.g., Diener & Emmons, 1984; Gotlib & Meyer,1986; Watson, 1988) and with evidence that each of thesedimensions has different neurophysiological substrates (Baker,Morse, & Sherman, 1987; Wise, 1988).

The rwo-dimensionai view of affect can be translated into a

two-dimensional view of affective predispositions in healthbehavior. In some health-behavior research, a two-dimen-sional view has been used to construct separate positive andnegative value x expectancy scales representing both beliefand affective elements of behavioral tendencies (Stacy, Dent,et al., 1990; Stacy, Widaman, & Marlatt, 1990). Although theseconstructs have been referred to as expectancies, they could beconsidered to be quasi-attitudinal constructs because theyinclude affective components. However, instead of merelypredicting attitude, in accord with traditional attitude perspec-tives (e.g., Fishbein & Ajzen, 1975), positive expectancies hada unique and superior predictive effect on behavioral inten-tions (Stacy, Widaman, & Marlatt, 1990). In addition, positiveand negative expectancies equally and strongly predictedattitude, yet positive and negative expectancies were onlymodestly intercorrelated. Other recent studies, using moretraditional assessments of outcome expectancies, also havefound that general factors of expectancies about positive andnegative outcomes were correlated only weakly, or nonsignifi-cantly (Leigh & Stacy, 1993; Stacy, MacKinnon, & Pentz,1993), providing further evidence that these constructs are notsimply cognitive elements of opposite poles of attitude.

Empirical support of the two-dimensional views of affectand expectancies suggests that traditional attitude constructsmay need to be complemented, if not replaced, by alternativeaffective and cognitive constructs in some areas of health-behavior research. It is possible that attitudes are merelyjudgments that people sometimes make but that do notrepresent the underlying source of judgments or behavior. Thissource may be somewhat less unitary or abstract than attitudeand may be better represented by sets of anticipated outcomesor expectancies. If so, then attitude, as traditionally definedand measured in the present studies, would not be expected tobe a strong predictor of behavior. Even though this perspectiveis only one of several different ways to interpret the lack ofstrong, general effects of attitude on health behavior, theprevious success of expectancies as long-term predictors ofsome health behaviors (Stacy, Newcomb, & Bentler, 1991)suggests further that the expectancy framework may be aviable complement (or alternative) to traditional attitudeframeworks.

Although alternatives to attitude may be needed to moreadequately reflect affective and cognitive influences on healthbehavior, attitude was predictive of health behavior in two ofthe models we analyzed (college alcohol use and communitymarijuana use). Past behavior was a nonspurious predictor ofattitude in two other models (adolescent smoking and collegebinge eating). A third interpretation of the pattern of results isthat the nature of the specific behaviors, the populationsampled, or the time interval used in these models may havecontributed to the obtained differences. In other words,attitudes may be more generally important than we found, butsomething correlated with the particular analyzed modelmoderated the predictive effects of attitude on health behavior.Although it is unclear what specific aspects of the populationsor the health behaviors could have acted as moderators ofattitude-behavior relations in the present studies, we offersome speculations about potential moderators. It is possible


that alcohol and marijuana use in mostly nonproblematicpopulations (college and community populations) are undergreater volitional control; for most people sampled from thesepopulations, these behaviors probably are not particularlyaddictive or compulsive. On the other hand, the literature onbinge eating and smoking suggests that these behaviors areoften quite compulsive or addictive in quality, even amongotherwise nonproblematic populations (e.g., college or highschool students). Alcohol use in groups of drunk-drivingoffenders is by definition more problematic, and is likely to bemore compulsive or addictive than alcohol use among collegestudents. This framework would suggest that, among moreproblematic samples or behaviors, previous behavior (as anindicator of habit) would be a very strong predictor of futurebehavior, overriding any potential effects of attitudes. In fact,in the present models, past behavior was a stronger predictorof future behavior in the four instances in which attitude wasnot predictive, compared with the two instances in whichattitude was predictive of health behavior. It also is possiblethat when behavior is particularly stable over time, it has agreater potential to influence attitudes, as found in our bingeeating and smoking models. This effect could occur throughstable and repeated self-perceptions (Bern, 1978) or otherprocesses (Stacy et al., 1991).

Another feature of the models we analyzed that maymoderate the predictive effects of attitude is the time intervalused to evaluate attitude-behavior relations. In several theo-ries, such as self-perception, cognitive dissonance, and attitudeaccessibility, the influence of attitudes and behavior on oneanother may be quite temporary and may not be captured bythe moderate to extremely long time intervals assessed in ourstudies. However, there was no apparent correlation betweenstrength of attitude-behavior relations in our studies and timeinterval. In fact, one of the only models to show a significantprediction of health behavior by attitudes was in the longeststudy, in which marijuana attitudes predicted marijuana useover a 9-year interval. Nevertheless, it is possible that muchshorter time intervals (e.g., days instead of weeks, months, oryears) would show a quite different pattern of findings.Although findings from very short interval studies would be ofgreat theoretical interest, their applied importance is notreadily apparent. Most attitude-change interventions in healthbehavior seek to make long-term changes in behavior.

Further research may delineate more specifically whatfeatures of the population and specific behavior are respon-sible for differences in attitude-behavior relations in health.The potential moderators mentioned earlier should be consid-ered for use in future research on attitudes in health behavior.The investigation of such moderators may reveal that attitudeshave a relevant degree of predictive utility under certain sets ofspecific conditions. If so, then developers of prevention pro-grams may be better able to determine the conditions underwhich an attitude-change program may be most successful. Onthe other hand, future research may confirm the alternativeview that attitudes are too limited in their predictive effects tobe useful in health-behavior interventions. Such findingswould provide further motivation for the use of alternativeaffective and cognitive constructs.

References

Ajzen, I., & Fishbein, M. (1977). Attitude-behavior relations: Atheoretical analysis and review of empirical research. PsychologicalBulletin, 84, 888-918.

Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predictingsocial behavior. Englewood Cliffs, NJ: Prentice Hall.

Ajzen, I., & Madden, T. J. (1986). Prediction of goal-directedbehavior: Attitudes, intentions, and perceived behavioral control.Journal of Experimental Social Psychology, 22, 453—474.

Akers, R. L. (1992). Drugs, alcohol, and society: Social structure, process,and policy. Belmont, CA: Wadsworth.

Akers, R. L., Krohn, M. D., Lanza-Kaduce, L., & Radosevich, M.(1979). Social learning and deviant behavior: A specific test of ageneral theory. American Sociological Review, 44, 636-655.

American Psychiatric Association. (1987). Diagnostic and statisticalmanual of mental disorders (3rd ed., rev.). Washington, DC: Author.

Baker, T. B., Morse, E., & Sherman, J. E. (1987). The motivation touse drugs: A psychobiological analysis of urges. In P. C. Rivers(Ed.), The Nebraska symposium on motivation: Alcohol use and abuse(pp. 257-323). Lincoln: University of Nebraska Press.

Bauman, K. E., & Dent, C. W. (1982). Influence of an objectivemeasure on self-reports of behavior. Journal of Applied Psychology,67, 623-628.

Bern, D. J. (1978). Self-perception theory. In L. Berkowitz (Ed.),Cognitive theories in social psychology, (pp. 221-282). San Diego, CA:Academic Press.

Ben-Tovim, D. I., & Walker, M. K. (1991). The development of theBen-Tovim Walker Body Attitude Questionnaire (BAQ), a newmeasure of women's attitudes towards their own bodies. Psychologi-cal Medicine, 21, 775-784.

Bentler, P. M. (1989). EQS structural equations program manual. LosAngeles: BMDP Statistical Software, Inc.

Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodnessof fit in the analysis of covariance structures. Psychological Bulletin,88, 588-606.

Bentler, P. M., & Speckart, G. (1979). Models of attitude-behaviorrelations. Psychological Review, 86, 452^64.

Bentler, P. M., & Speckart, G. (1981). Attitudes "cause" behaviors: Astructural equation analysis. Journal of'Personality and Social Psychol-ogy, 40, 226-238.

Biglan, A., Gallison, C., Ary, D., & Thompson, R. (1985). Expired aircarbon monoxide and saliva thiocyanate: Relationships to self-reports of marijuana and cigarette smoking. Addictive Behaviors, 10,137-144.

Bollen, K. A. (1989). Structural equations with latent variables. NewYork: Wiley.

Brunswick, A. F., & Messeri, P. (1984). Causal factors in onset ofadolescents' cigarette smoking: A prospective study of urban Blackyouth. Advances in Alcohol and Substance Abuse, 3, 35-52.

Bruvold, W. H. (1990). A meta-analysis of the California school-basedrisk reduction program. Journal of Drug Education, 20, 139-152.

Charlton, A., & Blair, V. (1989). Predicting the onset of smoking inboys and girls. Social Science and Medicine, 29, 813-818.

Corey, S. M. (1937). Professed attitudes and actual behavior. TheJournal of Educational Psychology, 28, 271-280.

Diener, E., & Emmons, R. A. (1984). The independence of positiveand negative affect. Journal of Personality and Social Psychology, 47,1105-1117.

Evans, R. I., Hansen, W. B., & Mittelmark, M. B. (1977). Increasingthe validity of self-reports of smoking behavior in children. Journalof Applied Psychology, 62, 521-523.

Farquhar, J. W. (1991). The Stanford cardiovascular disease preven-tion programs. Annals of the New York Academy of Sciences, 623,327-331.


Fazio, R. H., & Williams, C. J. (1986). Attitude accessibility as amoderator of the attitude-perception and attitude-behavior rela-tions: An investigation of the 1984 presidential election. Journal ofPersonality and Social Psychology, 51, 505-514.

Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior:An introduction to theory and research. Reading, MA: Addison-Wesley.

Flay, B. R. (1987). Mass media and smoking cessation: A criticalreview. American Journal of Public Health, 77, 153-160.

Flay, B. R., d'Avernas, J. R., Best, J., Kersell, M. W., & Ryan, K. B.(1983). Cigarette smoking: Why young people do it and ways ofpreventing it. In P. McGrath & P. Firestone (Eds.), Pediatric andadolescent behavioral medicine (pp. 132-183). New York: Springer-Verlag.

Flay, B. R., Ryan, K. B., Best, J. A., Brown, K. S., Kersell, M. W.,d'Avernas, J. R., & Zanna, M. P. (1985). Are social-psychologicalsmoking prevention programs effective? The Waterloo Study. Jour-nal of Behavioral Medicine, 8, 37-59.

Gorsuch, R. L. (1983). Factor analysis (2nd ed.). Hillsdale, NJ:Erlbaum.

Gotlib, I. H., & Meyer, J. P. (1986). Factor analysis of the multipleaffect adjective check list: A separation of positive and negativeaffect. Journal of Personality and Social Psychology, 50, 1161-1165.

Hawkins, R. C., II, & Clement, P. F. (1980). Development andconstruct validation of a self-report measure of binge eating tenden-cies. Addictive Behaviors, 5, 219-226.

Huba, G. J., & Harlow, L. L. (1987). Robust structural equationmodels: Implications for developmental psychology. Child Develop-ment, 58, 147-166.

Jarvis, M. J., Goddard, E., & McNeil, A. (1990). Do attitudes predictuptake of smoking in teenagers? Case not proven. Social Science andMedicine, 31, 997-1000.

Johnson, V. (1988). Adolescent alcohol and marijuana use: A longitu-dinal assessment of a social learning perspective. American Journalof Drug and Alcohol Abuse, 14, 419-439.

Jones, E. E., & Sigall, H. (1971). The bogus pipeline: A new paradigmfor measuring affect and attitude. Psychological Bulletin, 76, 349-364.

Joreskog, K. G. (1979). Statistical estimation of structural models inlongitudinal-developmental investigations. In J. R. Nesselroade &P. B. Baltes (Eds.), Longitudinal research in the study of behavior anddevelopment (pp. 303-351). San Diego, CA: Academic Press.

Kahle, L. R., & Herman, J. J. (1979). Attitudes cause behavior: Across-lagged panel analysis. Journal of Personality and Social Psychol-ogy, 37, 315-321.

Katzman, M. A., & Wolchik, S. A. (1984). Bulimia and binge eating incollege women: A comparison of personality and behavioral charac-teristics. Journal of Consulting and Clinical Psychology, 52, 423-428.

Kelman, H. C. (1974). Attitudes are alive and well and gainfullyemployed in the sphere of action. American Psychologist, 29, 310-324.

LaPiere, R. T. (1934). Attitudes vs. actions. Social Forces, 13, 230-237.Leigh, B. C., & Stacy, A. W. (1993). Alcohol outcome expectancies:

Scale construction and predictive utility in higher-order confirma-tory models. Psychological Assessment, 5, 216-229.

Leventhal, H., & Cleary, P. D. (1980). The smoking problem: A reviewof research and theory in behavioral risk modification. PsychologicalBulletin, 88, 370-408.

Madden, T. J., Ellen, P. S., & Ajzen, I. (1992). A comparison of thetheory of planned behavior and the theory of reasoned action.Personality and Social Psychology Bulletin, 18, 3-9.

Marsh, H. W., Balla, J. R., & McDonald, R. P. (1988). Goodness-of-fitindices in confirmatory factor analysis: The effect of sample size.Psychological Bulletin, 103, 391-410.

Murray, D. M., & Perry, C. L. (1987). The measurement of substanceuse among adolescents. When is the "bogus pipeline" methodneeded? Addictive Behaviors, 12, 225-233.

National Cancer Institute. (1986). Smoking, tobacco, and cancerprogram: 1985 report (NIH publication No. 86-2687). Washington,DC: National Institute of Health.

Newcomb, M. D., & Bentler, P. M. (1988). Consequences of adolescentdrug use: Impact on the lives of young adults. Newbury Park, CA:Sage.

Pechacek, T. F., Murray, D. M., Luepker, R. V., Mittelmark, M. B.,Johnson, C. A., & Schultz, J. M. (1984). Measurement of adolescentsmoking behavior: Rationale and methods. Journal of BehavioralMedicine, 7, 123-140.

Pentz, M. A., Dwyer, J. H., MacKinnon, D. P., Flay, B. R., Hansen, W.B., Wang, E. Y. I., & Johnson, C. A. (1989). A multi-community trialfor primary prevention of adolescent drug abuse: Effects on drug useprevalence. Journal of the American Medical Association, 261, 3259-3266.

Polivy, J., & Herman, C. P. (1985). Dieting and binging: A causalanalysis. American Psychologist, 40, 193-201.

Polivy, J., Herman, C. P., Olmstead, M. P., & Jazwinski, C. (1984).Restraint and binge eating. In R. C. Hawkins II, W, J. Fremouw, &P. F. Clement (Eds.), The binge-purge syndrome: Diagnosis, treat-ment, and research (pp. 104-122). New York: Springer.

Raden, D. (1985). Strength-related attitude dimensions. Social Psycho-logical Quarterly, 48, 312-330.

Rindskopf, D. (1984). Structural equation models: Empirical identifi-cation, Heywood cases, and related problems. Sociological Methods& Research, 13, 109-119.

Rogosa, D. (1980). A critique of cross-lagged correlation. PsychologicalBulletin, 88, 245-258.

Rosenberg, M. J., & Hovland, C. I., (1960). Cognitive, affective, andbehavioral components of attitudes. In C. I. Hovland & M. J.Rosenberg (Eds.), Attitude organization and change (pp. 1-14). NewHaven, CT: Yale University Press.

Schlegel, R. P., Manske, S. R., & d'Avernas, J. R. (1985). Alcohol anddrug use in young adults: Selected findings in a longitudinal study.Bulletin of the Society of Psychologists in Addictive Behaviors, 4,213-225.

Shapiro, A., & Browne, M. W. (1987). Analysis of covariance struc-tures under elliptical distributions. Journal of the American StatisticalAssociation, 82, 1092-1097.

Snow, W. H., Gilchrist, L. D., & Schinke, S. P. (1985). A critique ofprogress in adolescent smoking prevention. Children and YouthServices Review, 7, 1-19.

Snyder, M. (1974). The self-monitoring of expressive behavior. Journalof Personality and Social Psychology, 30, 526-537.

Speckart, G., & Bentler, P. M. (1982). Application of attitude-behavior models to varied content domains. Academic PsychologyBulletin, 4, 453-465.

Spencer, J. A., & Fremouw, W. J. (1979). Binge eating as a function ofrestraint and weight classification. Journal of Abnormal Psychology,88, 262-267.

Stacy, A. W., Dent, C. W., Sussman, S., Raynor, A., Burton, D., &Flay, B. R. (1990). Expectancy accessibility and the influence ofoutcome expectancies on adolescent smokeless tobacco use. Journalof Applied Social Psychology, 20, 802-817.

Stacy, A. W., Flay, B. R., Sussman, S., Brown, S., Santi, S., & Best, J. A.(1990). Convergent validity of alternative self-report indices ofsmoking among adolescents. Psychological Assessment, 2, 442-446.

Stacy, A. W., MacKinnon, D. P., & Pentz, M. A. (1993). Generalityand specificity in health behavior: Application to warning label and


social influence expectancies. Journal of Applied Psychology, 78,611-627.

Stacy, A. W., Newcomb, M. D., & Bentler, P. M. (1991). Cognitivemotivation and problem drug use: A nine-year longitudinal study.Journal of Abnormal Psychology, 100, 502-515.

Stacy, A. W., & Saetermoe, C. (1987, May). The validity of scales ofdieting and binge eating. Paper presented at the annual meeting ofthe Western Psychological Association, Long Beach, CA.

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

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

Thurstone, L. L. (1931). The measurement of social attitudes. Journalof Abnormal and Social Psychology, 26, 249-269.

Watson, D. (1988). The vicissitudes of mood measurement: Effects ofvarying descriptors, time frames, and response formats on measuresof positive and negative affect. Journal of Personality and SocialPsychology, 55, 128-141.

Wicker, A. W. (1971). An examination of the "other variables"explanation of attitude-behavior inconsistency. Journal of Personal-ity and Social Psychology, 19, 18-30.

Wise, R. A. (1988). The neurobiology of craving: Implications for theunderstanding and treatment of addiction. Journal of AbnormalPsychology, 97, 118-132.

Wyer, R. S., Jr., & Srull, T. K. (1989). Memory and cognition in its socialcontext. Hillsdale, NJ: Erlbaum.

AMERICAN PSYCHOLOGICAL ASSOCIATIONSUBSCRIPTION CLAIMS INFORMATION Today's Date:_

We provide this form to assist members, institutions, and nonmember individuals with any subscription problems. With theappropriate information we can begin a resolution. If you use the services of an agent, please do NOT duplicate claims throughthem and directly to us. PLEASE PRINT CLEARLY AND IN INK IF POSSIBLE.

PRINT FULL NAME OR KEY NAME OF INSTITUTION

ADDRESS

MEMBER ORCUSTOMERNUMBER(MAYBEFOUND ON ANYPASTISSUE LABEL)

DATE YOUR ORDER WAS MAILED (OR PHONED)

_CHECK ___CHARGECHECIWCARD CLEARED DATE:_

STATE/COUNTRY

YOUR NAME AND PHONE NUMBER

TITLE

(If possible, send a copy, front and back, of your cancelled check to help us in our researchof your claim.)

ISSUES: MISSING DAMAGED

VOLUME OR YEAR NUMBER OR MONTH

Thank you. Once a claim is received and resolved, delivery of replacement issues routinely takes 4-6 weeks.

—————— (TO BE FILLED OUT BY APA STAFF) ̂ ——

DATE RECEIVED:.ACTION TAKEN: _STAFF NAME:

DATE OF ACTION: _INV.NO.&DATE:LABEL NO. & DATE:

Send this form to APA Subscription Claims, 750 First Street, NE, Washington, DC 20002-4242

PLEASE DO NOT REMOVE. A PHOTOCOPY MAY BE USED.

Documents

Attitudes and Health Behavior in Diverse Populations ...people.oregonstate.edu/~flayb/MY PUBLICATIONS/Multiple behaviors... · The affect-based definition of attitude, in which an