Read Stacy%20et%20al%2094.pdf text version

Health Psychology 1994, Vol. 13, No. 1,73-85

Copyright 1994 by the American Psychological Association, Inc., and the 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 drunk driving) were studied prospectively in 5 different groups of subjects. Associations between attitudes toward these behaviors and the behaviors themselves were investigated over at least 2 waves of measurement. Findings revealed that attitudes predicted behavior nonspuriously in 2 instances: alcohol use and marijuana use. Attitudes did not predict drunk driving, binge eating, or smoking behaviors. Past behavior predicted attitude in the domains of binge eating and smoking, but not in the domains of alcohol use, drunk driving, or marijuana use. The results are discussed in terms of several alternative approaches that have implications for interventions that attempt to influence health behavior through attitude change. Key Words: attitudes, health, behavior, longitudinal, expectancy, drugs

Programs designed to promote changes in health behavior frequently include components that attempt to change attitudes concerning the targeted behavior. Sometimes the focus on attitude change is an explicit component, in which an individual's evaluations of the behavior are targeted. In other cases, as in many educational media campaigns (e.g., Flay, 1987), several key or "salient" beliefs are the focus of the strategy, which more implicitly attempts attitude change by first trying to change beliefs (cf. Ajzen & Fishbein, 1980). Programs that provide information about the harmful consequences of smoking, the dangers of drunk driving, or the necessity of a balanced diet often represent, implicitly or explicitly, an attempt to change behavior through attitude change. These programs may focus on harmful outcomes of a negative health behavior, attempting to make personal evaluations (attitude) of the behavior more negative, or on beneficial outcomes of a positive health behavior, attempting to make evaluations of the healthy behavior more positive. Because of the widespread use and potential of attitude concepts in health-behavior research and interventions, it is important to document the predictive impact of this construct and to integrate theories of health behavior with empirical findings. The rationale for attitudinal approaches in health-behavior interventions has been strengthened by findings in basic research documenting the association between attitudes and behavior (Bentler & Speckart, 1981; Fazio & Williams, 1986;

Alan W. Stacy, Department of Preventive Medicine, Institute for Prevention Research, University of Southern California, Peter M. Bentler, Department of Psychology, University of California, Los Angeles; Brian R. Flay, Prevention Research Center, School of Public Health, University of Illinois, Chicago. Support for this research was provided by Grant DA01070-20 from the National Institute on Drug Abuse. The views expressed herein are ours and do not necessarily reflect the opinions of the Public Health Service. Correspondence regarding this article should be addressed to Alan W. Stacy, Institute for Prevention Research, University of Southern California, 1000 S. Fremont, #641, Alhambra, California 91803-1358.


Fishbein & Ajzen, 1975; Speckart & Bentler, 1982). However, comprehensive evaluations of the predictive impact of attitudes on health behavior have been rare, and the case for attitude as a strong natural precursor to behavior may not be as strong as some of the basic research would suggest. Our goal was to investigate the predictive precedence of attitude and health behavior over time, so that the usefulness of the attitude construct in health can be more thoroughly examined. First, it is important to summarize briefly the predominant definitions of 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 a general evaluation of an object, and affect is considered the most essential aspect of attitude (also see Thurstone, 1931). In another dominant view, attitude is considered to be multidimensional, composed of correlated but somewhat independent affective, cognitive, and conative components (Rosenberg & Hovland, 1960). The affect-based definition of attitude, in which an overall evaluation of an object or a behavior is emphasized, is probably the more common one in present research in social psychology and health, and this is the definition we use in this article. A focus on this definition of attitude does not imply that cognitive factors are not important in health behavior, but instead suggests that cognitive variables are better construed as separate but related constructs (e.g., Ajzen & Fishbein, 1980; Stacy, Widaman, & Marlatt, 1990). Attitudinal research has frequently focused on the strength or nature of attitude-behavior relations, attempting to shed light on the predictive validity of attitude concepts. Although early research often found poor correlations between attitude and behavior (e.g., Corey, 1937; LaPiere, 1934; for review, see Wicker, 1971), Fishbein and Ajzen (1975; Ajzen & Fishbein, 1980) suggested that relations between attitudes and behavior would be stronger when measurements of these variables correspond highly in certain properties. Specifically, Ajzen and



Fishbein's (1977) review of the literature identified strong attitude-behavior associations across a number of studies in which at least the target (object) and action (e.g., drinking) elements of measurement corresponded highly. Fishbein and Ajzen's (1975) approach is consistent with the position that attitudes have predictive priority over behavior (e.g., Bentler & Speckart, 1981). Other positions relevant to health behavior include the view that behavior causes attitude (Bern, 1978), attitudes and behavior influence each other reciprocally (Kelman, 1974), and the influence of attitude on behavior depends strongly on moderator variables, such as individual differences, in the accessibility of attitudes or beliefs from memory (Fazio & Williams, 1986; Stacy, Dent, et al., 1990) or in personality characteristics (e.g., Snyder, 1974). Although these latter perspectives have made contributions to the study of moderators of attitude-behavior relations, most health-behavior campaigns focus on attitude and behavior change rather than on attitude-moderator change. Thus, at least in health-behavior research, it is important to determine the precedence of attitudes and behavior as direct-effect predictors of one another. Surprisingly, questions about the predictive priority of attitudes and behavior have been addressed rarely in health-behavior research.

Alcohol Use and Drunk Driving Existing evidence about attitude-behavior relations in alcohol-use research can be considered mixed. Kahle and Herman (1979) found that attitudes showed predictive predominance over behavior in a cross-lagged panel correlational investigation of alcohol consumption over a 2-month period (also see Schlegel, Manske, & d'Avernas, 1985), but this type of statistical analysis has been criticized on several grounds (Rogosa, 1980). In a structural equation analysis, Bentler and Speckart (1979; also see Stacy, Widaman, & Marlatt, 1990) found evidence that attitude toward alcohol use predicted drinking behavior 2 weeks later, although they did not perform an examination of predictive precedence. In a 3-year prospective study, Johnson (1988) found no evidence that attitudes about drinking and drinking behavior predicted one another over time, in either of the possible directions. No studies have investigated the predictive precedence of attitudes and behavior regarding alcohol-related problem behaviors, such as driving under the influence of alcohol (DUI). However, the call for attitude change regarding alcohol use frequently accompanies discussions of the social costs of this and other problems associated with alcohol use. Illicit Drug Use

Attitudes and Specific Health Behaviors Smoking

The influence of attitudes on smoking onset and cessation is important because attitude-change strategies are frequently used 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 elements of successful interventions in this area (Flay, 1987; Snow, Gilchrist, & Schinke, 1985), theoretically sound attitudechange strategies are in principle quite amenable to mass media efforts and school-based interventions. Because most adult smokers begin to smoke in adolescence (Flay, d'Avernas, Best, Kersell, & Ryan, 1983; Leventhal & Cleary, 1980), this age group seems particularly important for research and intervention. Charlton and Blair (1989; but see Jarvis, Goddard, & McNeil, 1990) reported evidence that attitudes among initial nonsmokers predicted adolescent smoking behavior prospectively, but they did not investigate the predictive precedence of attitude and behavior over time. In other instances, attitudes that do not correspond in level of specificity to this behavioral target and action (smoking) have been studied in smoking research, including attitudes toward self or toward health (e.g., Brunswick & Messeri, 1984). Consistent with Ajzen and Fishbein's (1977) predictions and findings, a lack of correspondence in the target or action element of attitudes and behavior has led to low attitude-behavior relationships. Despite the widespread use of attitude concepts in both interventions and theories of adolescent smoking, no studies have investigated the predictive precedence of attitudes toward smoking and smoking behavior in this population.

As with tobacco and alcohol use, many intervention efforts attempt some form of direct or indirect attitude change in the prevention and treatment of drug abuse (e.g., Pentz et al., 1989). Although attitudes appear to be predictive of certain types of illicit drug use (e.g., Akers, Krohn, Lanza-Kaduce, & Radosevich, 1979), few studies have examined whether this prediction is as strong as the prediction of attitudes by drug use. If the prediction of attitudes by drug use is much stronger than the prediction of drug use by attitudes, then it may be the case that behavior must change before attitude changes. A predictive predominance of behavior over attitude would have implications for the focus of interventions. The only existing evidence concerning the predictive precedence of attitudes and illicit drug use was reported by Johnson (1988) and by Schlegel et al. (1985). Johnson found that attitude and marijuana use did not predict one another over time, and that only the stability (autoregressive) effects of the same construct on itself over time were significant. Schlegel et al. reported only preliminary information about predictive precedence, as evaluated in a cross-lagged correlational design of marijuana attitudes and behavior. Binge Eating During the past two decades, binge eating has appeared to increase in frequency among young women (Polivy & Herman, 1985). Although the behavior of binge eating may be part of the larger eating disorder of bulimia, people who binge eat may vary continuously in their extent of disordered eating behavior and concomitant symptoms. Research investigating this behavior is important because of a number of negative health consequences of the behavior (e.g., Hawkins & Clement, 1980), whether or not a clear, clinically relevant case of

ATTITUDES AND HEALTH BEHAVIOR bulimia is evident. Even when binge eating is not apparently severe, the typical diet associated with binge eating is high in fat and is likely to cause health consequences if the diet continues for an extended period of time. A number of studies have found that cognitive factors more consistently trigger binge-eating behaviors than do physiological variables, such as caloric intake (e.g., Polivy, Herman, Olmstead, & Jazwinski, 1984; Spencer & Fremouw, 1979), and that certain types of attitudes are associated with this behavior. However, the only types of attitudes that have been investigated are attitudes about one's body (for review, see BenTovim & Walker, 1991). Attitudes toward the target behavior (e.g., Fishbein & Ajzen, 1975) have not been investigated as a predictor of binge eating, so the degree to which binge eating is under attitudinal control is unknown at present. Thus, although attitudinal approaches to primary prevention of binge eating among adolescents and young adults may be theoretically feasible, the absence of empirical evidence showing a relation between attitudes and behavior in this domain dampens the potential enthusiasm for such an approach.



Little is known about the precise relations between attitude and the health behaviors we address in this article. When attitudes have been studied, attitude and behavior usually have not been measured prospectively over time in a way that would reveal 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 drug use behaviors and attitudes in five different groups sampled from populations in which these problem health behaviors are relevant. Although there are many other variables relevant to the study of attitude-behavior relations (e.g., Fazio & Williams, 1986; Madden, Ellen, & Ajzen, 1992; Stacy, Widaman, & Marlatt, 1990), the only relevant variables available across each of the present studies were measures of attitude and behavior. As stated earlier, the investigation of the predictive relations between attitude and health behavior is important in its own right, especially when a range of different health behaviors are investigated prospectively. Furthermore, many health-promotion interventions using attitude-change strategies have assumed implicitly that attitudes have direct effects on behavior, consistent with the predictive models we evaluated.

med fit index (NNFI; Bentler & Bonett, 1980), which has been found to be among the most robust indexes of fit regarding sample-size biases (Marsh, Balla, & McDonald, 1988). We also report the normed fit index (NFI; Bentler & Bonett, 1980). In each of the studies, measures of attitude and behavior corresponded specifically in at least the target (e.g., cigarettes) and action (e.g., smoking) elements of measurement, which have been found to be most important in the study of attitude-behavior relations (Ajzen & Fishbein, 1977). The context element (e.g., place where smoking occurs) remained quite general in all of the studies, but corresponded across measures of attitude and behavior. The time element (e.g., smoking in next month) corresponded very specifically in the three short-term studies and more generally in the two longer-term studies. For each behavioral domain we studied, only subjects who indicated they had engaged in the relevant behavior were used in the analysis. Thus, the different samples were equated for having at least minimal experience with the behavior. Equating the samples in this way reduced the possibility that attitude-behavior relations differed across samples because of differences in relevant experience--a likely moderator of attitude effects. A number of findings in basic research show that this moderator confound is likely, because subjects who differ in direct experience with an object typically show different strengths of attitude-behavior relations (for review, see Raden, 1985). In these studies, the subjects with direct experience generally showed stronger levels of prediction of behavior by attitudes. Clearly, some type of control of level of experience is necessary if results from different samples with varying levels of experience are to be compared. The subjects in each study were predominantly middle class and White, but represented several different age and nationality groups.

Study 1: Binge Eating Method

Subjects. Respondents were 260 undergraduate female college students who participated in the study for extra credit in introductory psychology classes. Although 241 subjects completed all three waves of measurement and provided complete data, only the subjects who participated at all waves and who indicated they binge ate at some time in the past year (N = 144) were included in data analysis. We sampled only women because this problem health behavior generally is much more frequent in women than in men. Procedure. The survey was given at three different times, separated by 2-week intervals. Respondents completed questionnaires at predesignated times in a classroom, group-testing situation. Questionnaire identification numbers were used to allow for matching of questionnaires across waves; these numbers were self-generated by the subjects according to a previously pilot-tested protocol that ensures a high level of matches (95% or more) over a short period. The anonymity of the questionnaire was emphasized both by the administrator and in written statements in the questionnaire. Previous research has shown that if anonymity is guaranteed fully, valid responses to socially proscribed behaviors are likely even among young adolescents providing self-reports of drug use (Murray & Perry, 1987). Strong assurances of anonymity appear to be as effective as are bogus pipeline procedures in studies of consummatory behavior (Murray & Perry, 1987). Measures of behavior. The behavioral measures at each measurement asked respondents to report on their binge eating in the past 2 weeks. Before answering the binge-eating questions, respondents were asked to read a common definition of binge eating: "Binge eating is the rapid consumption of a large amount of food in a discrete period of time, usually less than two hours." For this behavior, 18 activities were

General Method

In each study reported below, we measured attitude and behavior prospectively and used covariance structure analysis (e.g., Bentler, 1989; Bentler & Speckart, 1981; Bollen, 1989) to investigate the predictive precedence of the two primary variables. We report results based on maximum likelihood estimation. This form of estimation has been found to be relatively robust to departures in multivariate normality (e.g., Huba & Harlow, 1987), but we also evaluated each of the covariance structure models using at least one other method of estimation. We evaluated the statistical significance of paths using chi-square difference tests and critical ratios; the overall goodness of fit of the models was evaluated primarily through use of the nonnor-


A. STACY, P. BENTLER, AND B. FLAY between same-wave factors at Tl and same-wave residual factor covariances at T2 and T3 also were freely estimated. We used the chi-square value for this initial model (x2 [110, N = 144] = 144.98,;? = .01) in model comparisons with morerestricted models. Thus, the paths in the initial model were tested for statistical significance by deleting paths one at a time from the initial model and evaluating the decrease in fit from the initial model in single-degree-of-freedom chi-square difference tests (cf. Bentler & Speckart, 1981). Using this procedure, all nonsignificant (p < .05) paths, factor covariances, and factor residual covariances were deleted from the initial model, deriving the final model. This final model (x2 [118, N = 144] = 154.97,;? = .01) fit the data well regarding practical indices of fit (NNFI = .99; NFI = .96). An alternative estimator appropriate for relatively small samples, using elliptical distribution theory (Shapiro & Browne, 1987), yielded identical results regarding significance tests of alternative paths. No single path, factor covariance, or factor residual covariance could be added to this final model to increase its fit significantly, nor could any single estimate in the final model be omitted without resulting in a significant decrease in model fit. In a more global test, the final model was not significantly different in fit from the initial model (x2 [8, N = 144] = 9.98, p > .05). The only significant cross-lagged path in the final model was from Tl behavior to T2 attitude. In addition, covariances between Tl attitude and behavior factors and between T3 attitude and behavior factor residuals were nonsignificant. The final model is shown in Figure 1, which presents the standardized structural parameter estimates and residual variances. Standardized factor loadings in the measurement model all were statistically significant (allps < .001) and were of adequate size, ranging from .71 to .96 for attitude and from .96 to .98 for binge-eating. These moderate to large loadings indicated that these indicators were measured with good internal consistency; they were well above the minimum loading sizes recommended prior to interpretation in factor analysis (e.g., Gorsuch, 1983).

assessed that characterize the behavior, consistent with criteria from the Diagnostic and Statistical Manual of Mental Disorders (3rd ed., rev.; Z>SM-///-rR; American Psychiatric Association, 1987) and frequently used definitions (Katzman & Wolchik, 1984; Polivy et al., 1984). These items were consistent with most criteria of binge eating based on these sources, but DSM-III-R criteria that refer more specifically to bulimia were omitted (e.g., dieting and purging). Thus, in our study, binge eating was used as a more general construct than bulimia per se. Most items were scored from 1 to 7, with frequency items anchored by not at all (1) and every day (7). The binge items assessed the number of days respondents: binge ate, ate large amounts of food while feeling compelled to do so, were out of control in their binge eating, hid binge eating from others, could not stop eating voluntarily, ate an extreme amount of a single type of food, consumed more than 1,200 calories per binge, ate much more than needed beyond satisfying hunger, ate much more than typical for body size, ate high-caloric and easily ingestible food, felt depressed after binge eating, kept eating until they thought they would explode, ate to the point of feeling sick, ate a large amount of food when not hungry, felt food controlled their life, and ate much more rapidly than most people ever eat. An additional item asked respondents how much food they ate very rapidly when they binge ate, coded from 1 (none) to 7 (a very large amount). A self-rating item (see Stacy, Widaman, Hays, & DiMatteo, 1985) assessed how respondents 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 three indicators of binge eating by randomly assigning each of the aforementioned binge items to one of the indicators, resulting in three indicators composed of simple sums of 6 items each. We conducted a preliminary study to evaluate the construct validity of these measures. In that study, the binge-eating measures were found to show adequate levels of item convergence and discrimination when thorough multitraitmultimethod procedures were used (Stacy & Saetermoe, 1987). For example, binge-eating indicators did not load on other behavioral factors (dieting and alcohol use), all binge-eating loadings were substantial (above .81), and none of the potential assessment method factors appeared to bias the results. Measures of attitude. Attitude at each time of measurement was assessed 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-point bipolar adjective items ranging from extremely good to extremely bad, extremely beneficial to extremely harmful, extremely rewarding to extremely punishing, extremely pleasant to extremely unpleasant, extremely enjoyable to extremely unenjoyable, and extremely favorable to extremely unfavorable. These measures are based on the recommendations of Ajzen and Fishbein (1980). Cross-lagged structural models. All of the models we tested used the same measurement model, in which the scale of measurement was determined by fixing Tl factor variances at 1.0 and imposing equality constraints on loadings of identical indicators of the same construct measured at the three different times. In addition, the measurement model specified covariances among unique scores of identical indicators measured across the three times. These procedures are consistent with those of Joreskog (1979).

Study 2: College Alcohol Use Method

Subjects. Respondents were 356 male and female undergraduate college students enrolled at a small West Coast university. Of the total, 334 respondents provided complete data and completed both waves of measurement. A subsample of respondents who indicated that they had drunk at least some alcohol in the previous year was selected for analysis (N = 270). The students received extra credit in an introductory-level psychology course for their participation in the study. These data were part of a larger study on alcohol use (Stacy, Widaman, & Marlatt, 1990). In the previous study, we did not examine a model in which behavior could predict attitude over time. Because the present cross-lagged structural model, which is parallel to the other models analyzed in this article, was not analyzed in the previous study, it is appropriate to report these analyses here. Procedure. The procedure was identical to the one reported in Study 1, except that we assessed each subject at two different times, separated by a 4-week interval. Anonymity of responses was again


The initial, structurally saturated model specified all four possible predictive paths from Tl attitude and binge factors to T2 attitude and binge factors. An analogous set of four paths was estimated among T2 and T3 attitude and binge factors and among Tl and T3 attitude and binge factors. Covariances



Binge Eating Time 1


Binge Eating Time 2

Binge Eating Time 3

Attitude Time 1



/ Attitude Time 3

Figure 1. Final structural equation model and standardized maximum likelihood estimates for binge eating. (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 residuals are depicted with double-headed arrows. Significance levels are based on critical ratios on unstandardized maximum likelihood estimates [*p < .05; **p < .01; ***p < .001].)

thoroughly emphasized. Three indicators were developed for each latent variable. Measures. Measures of attitude were identical to those outlined in Study 1, but were worded in terms of alcohol use. That is, subjects responded to identical semantic differential attitude items but reported attitudes in terms of "your drinking alcohol in the next four weeks." Alcohol-use measures were constructed to assess the general extent (i.e., both quantity and frequency) of alcohol use during the past 4 weeks. The first indicator of alcohol use was the sum of two 7-point scales, closely modeled after scales used by Bentler and Speckart (1979,1981). Both scales assessed frequency of use; one had endpoints of every day and not at all, and the second scale had endpoints of extremely frequently and never. The second indicator used an 8-point rating scale on which each respondent was instructed to classify himself or herself into 1 of 8 drinker categories (very heavy drinker, heavy drinker, fairly heavy drinker, moderate drinker, fairly light drinker, light drinker, very light drinker, and nondrinker). The third indicator was a quantity-frequency, or QF, index in which quantity and frequency of typical use (number of drinks consumed in previous month, based on typical quantity consumed on days the subject drank) was summed with greatest use (greatest number of drinks the subject consumed multiplied by the number of days the subject drank this greatest amount). A thorough multitrait-multimethod assessment of similar alcohol indicators has demonstrated their convergent and discriminant validity (Stacy et ah, 1985). In addition, Stacy et al. found evidence that method effects, such as general self-report biases, were not strong influences on this type of data.

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

Results The initial, structurally saturated model specified all four possible predictive paths from Tl attitude and alcohol factors to T2 attitude and alcohol factors. Covariances between same-wave factors at Tl and same-wave residual factor covariances at T2 also were freely estimated. We used the chi-square value for this initial model (x2 [46, N = 270] = 73.40, p < .01) in model comparisons with more-restricted models, using the same model testing and parameter trimming procedures as reported in Study 1. Only one path, from Tl behavior to T2 attitude, was nonsignificant using this procedure, and this path was eliminated to derive the final model. The final model (x2 [47, N = 270] = 74.98, p < .01) fit the data well according to practical indices of fit (NNFI = .98; NFI = .97). Elliptical theory and arbitrary distribution theory estimation (Bentler, 1989) led to the same final model. The final model did not differ significantly from the initial model. The standardized structural parameter estimates for the final model are provided in Figure 2. Standardized factor loadings ranged from


A. STACY, P. BENTLER, AND B. FLAY favorable): "Your attitude toward or feeling about your driving after drinking in the next month is?" This item is in the form of one of the alternative methods of measuring attitude suggested by Ajzen and Fishbein (1980). DUI behavior was measured by three items. One item asked respondents how often they drove after drinking alcohol in the past month, coded from 1 (not at all) to 7 (every day). A second item asked respondents how often they drank at least three drinks right before driving in the past month, coded on the same 7-point scale. The third item asked respondents to rate what kind of drinker they were when 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 attitudebehavior relations (DUI and alcohol use per se) were examined in two different series of models. For the alcohol-use domain, the models we tested used the same model construction and evaluation procedures as in Studies 1 and 2. For the DUI domain, the behavior factor at Tl and T2 represented three indicators, as with the aforementioned alcohol factor, but the attitude factors at both times were composed of a single indicator, consistent with a standard path-analytic specification (e.g., Bentler, 1989). The model-evaluation procedure for the DUI domain was the same as the procedure used in the alcohol-use domain and in Studies 1 and 2. Results Alcohol Use. The chi-square values for the initial, saturated structural model (x2 [46, N = 164] = 53.44, p = .21) were employed in model comparisons with more-restricted models, using the same model-testing and parameter-trimming procedures as reported in the previous studies. The two cross-lagged paths and the covariance between the factor residuals at T2 were found to be nonsignificant. These three estimates were omitted to derive the final model. The final model (x2 [49, N = 164] = 57.68,p = .19) fit the data very well using practical indices of fit (NNFI = .99; NFI = .97). The final model did not differ significantly from the initial model in terms of model fit, and the significance of paths did not change when we evaluated the models with alternative (elliptical and arbitrary distribution) estimators. Standardized structural parameter estimates for the final model are reported in Figure 3. Standardized factor loadings ranged from .81 to .97 for attitude and from .89 to .96 for alcohol use (all/?s < .001). Driving under the influence of alcohol. The initial model was structurally 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 comparisons with more-restricted models, but more-restricted models could not be estimated, because of linear dependencies in estimation. Even when a model is mathematically identified, such dependencies can be a product of "empirical underidentification" (resulting from various peculiarities in the data; Rindskopf, 1984). However, the structurally saturated model did not contain any dependencies or estimation problems, and the statistical significance of parameter estimates therefore could be evaluated on the basis of z-tests from this model (instead of model-trimming procedures). Neither of the cross-lagged paths, from attitude to behavior or from behavior to attitude, was significant at p < .05 according to these tests. Standardized structural parameter estimates from this model are provided in Figure 4. Standardized factor loadings on DUI behavior ranged from .70 to .95 and each was


Figure 2. Final structural equation model and standardized maximum likelihood estimates for college alcohol use. (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 residuals are depicted with double-headed arrows. Significance levels are based on critical ratios on unstandardized maximum likelihood estimates [*p < .05;**/> < .01; "**p < .001].)

.82 to .86 for attitude and from .8 I to .93 for alcohol use (all ps < .001).

Study 3: Alcohol Use and Drunk Driving Among DUI Offenders

Method Subjects. Respondents were 175 men who had been convicted of driving under the influence (DUI) of alcohol and who were participating in an educational program mandated by the state of California. Because all of the subjects drank alcohol and had engaged in DUI in the recent past, no subjects were removed from the analysis. Out of this sample, 164 respondents completed both waves of measurement. Respondents were paid $5.00 to complete the questionnaire, which was completely voluntary and anonymous. Procedure. The procedure was identical to the one described in Study 2. Subjects again were assessed at two different times, separated by a 4-week interval. No subjects completed the program before the final assessment. Respondents completed questionnaires on a voluntary basis after regular driver-education program meetings, in a classroom setting. Anonymity of responses was again thoroughly emphasized, but respondents were further assured that their participation in the study had nothing to do with the education program and that no one, including the program staff, had any way to discover their responses. Measures of alcohol use behavior and attitude. Three indicators were developed for each latent variable. Measures of attitude toward alcohol use were identical to those outlined in Study 2, except that one of three attitude indicators contained only one item because only five attitude items were used in the questionnaire. Alcohol-use measures included the three indicators used in Study 2, using the same frequency, rating, and QF methods of measurement; similar alcoholuse indicators were validated by Stacy et at. (1985). Measures of DUI behavior and DUI attitude. Attitude toward driving under the influence of alcohol was assessed with a single bipolar item that asked respondents to answer the following question on a scale coded from 1 (extremely unfavorable) to 7 (extremely



statistically significant (all ps < .001); because it was a single indicator, DUI attitude had no factor loadings. This model fit the data somewhat less well than did the preceding final models (NNFI = .92; NFI = .94). Use of the alternative estimators did not change the pattern of significance of structural paths. Study 4: Adolescent Smoking Method

Subjects. Respondents were 199 male and female llth-grade Canadian high-school students who indicated they had smoked at least once in their lives. These students constituted a subsample of a larger sample of students involved in a longitudinal study of adolescent smoking described by Flay et al. (1985). Procedure. We assessed each subject at two different times, separated by approximately 12 months. Respondents completed questionnaires in regular high school classrooms. Confidentiality, but not anonymity, of responses was emphasized. A number of studies have indicated that self-reports of cigarette use among adolescents are reasonably valid as long as certain safeguards are present during questionnaire administration (e.g., Bauman & Dent, 1982; Evans, Hansen, & Mittelmark, 1977; Murray & Perry, 1987; Pechacek et al., 1984). For example, Murray and Perry recommended that pipeline methods (cf. Jones & Sigall, 1971), which involve obtaining biological specimens, should be used to increase the validity of self-reports in adolescents when anonymity cannot be guaranteed. Thus, we used a pipeline procedure in which carbon monoxide (CO) measures were taken from each student to increase the validity of self-reports of cigarette use. Stacy, Flay, et al. (1990) found that this procedure yielded an adequate level of convergent validity between self-reported cigarette use and CO measurement in adolescents from the same population of subjects as sampled in this study.

DUI-Offender Drinking Driving Time 1


DUI-Offender DrinkingDrlvlng Time 2


Attitude Time 1


Attitude Time 2

Figure 4. Final structural equation model and standardized maximum likelihood estimates for DUI-offender drinking-driving. (Large circles designate latent constructs. Rectangles represent measuredvariable, or single-indicator, constructs. Small circles depict residual variances of constructs. Path coefficients are depicted with singleheaded arrows. Correlations between constructs or their residuals are shown with double-headed arrows. Significance levels are based on critical ratios on unstandardized maximum likelihood estimates ["p < .05; "p < .01; ***p < .001].) Measures of attitude. Measures of attitude were quite similar to the ones used in the previous studies. Attitude was assessed with three indicators that were each simple sums of two items. The items composing this scale were introduced with the statement "Smoking is": followed by six 5-point bipolar adjective items ranging from very beautiful to very ugly, very good to very bad, very clean to very dirty, very safe to very dangerous, very pleasant to very unpleasant, and very nice to very awful. Measures of smoking. Three different indicators of smoking were used. The first indicator was the smoking index used by Biglan, Gallison, Ary, and Thompson (1985), which is computed on the basis of the average of two items, including open-ended scales assessing last 24-hour use and past-7-day use. The other two indicators were proposed by the National Cancer Institute (NCI, 1986). One of these indicators was a closed-ended, 10-point scale of current smoking that combines frequency and quantity, asking respondents "How much do you currently smoke?" (I have never smoked = 1, Not at all in the last 12 months = 2, A few times in the last 12 months = 3, Usually once a month = 4, A few times each month = 5, Usually once a week = 6, A few times each week = 7, A few times most days = 8, About half a pack each day = 9, A pack or more each day = 10.). The second NCI indicator of smoking was a closed-ended item with 7 points, asking respondents "How frequently have you smoked cigarettes during the past 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 demonstrated the convergent and discriminant validity of these types of measures when used as continuous indexes (Stacy et al., 1985; Stacy, Flay, et al., 1990). Cross-lagged structural models. The model construction and evaluation procedures of Studies 1 and 2 were used in this study for the initial and subsequent models.

DUI-Offender Drinking Time 1

DUI-OTfender Drinking Time 2


Figure 3. Final structural equation model and standardized maximum likelihood estimates for driving under the influence (DTJI)offender alcohol use. (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 residuals are depicted with double-headed arrows. Significance levels are based on critical ratios on unstandardized maximum likelihood estimates ['p < .05; **p < .01; ***p < .001].)

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


A. STACY, P. BENTLER, AND B. FLAY tion in the larger study from representative junior high schools from Los Angeles County. Procedure. We assessed each subject at two different times, separated by approximately five years. Respondents completed a selfadministered questionnaire at both times of measurement. Although the data were not anonymous, the complete confidentiality of responses was emphasized to respondents, who were informed that their responses were protected by a federal certificate of confidentiality. Measures of attitude and behavior. The two measures of attitude asked respondents to indicate what they thought about smoking marijuana or taking PCP, on scales from 1 (terrible idea) to 5 (great idea). These measures were found to show an adequate degree of reliability in previous research as shown in high loadings on their respective factors (Bentler & Speckart, 1979). The measures of illicit drug-use behavior assessed the frequency of use of a variety of different 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 of different drugs, we restricted the primary analyses to marijuana use because of results from some preliminary analyses described below. Cross-lagged structural models. The model-evaluation procedures were the same as those used in the previous studies. However, the construction of the primary model was different because of some results from a preliminary analysis. As outlined above, attitude measures were available only for marijuana and PCP use. In the preliminary analysis, we attempted to model attitude-behavior relations between a general factor of drug-use attitudes, representing the

restricted models, using the procedures reported in the previous studies. One cross-lagged path, from Tl attitude to T2 smoking, was found to be nonsignificant; this path was omitted to derive the final model. The final model (x2 [47, N = 199] = 132.50, p < .001) fit the data well according to practical indices of fit (NNFI = .96; NFI = .95). The final model did not differ significantly from the initial model in terms of model fit. Standardized structural parameter estimates for the final model are reported in Figure 5. Standardized factor loadings ranged from .84 to .91 for attitude and from .90 to .97 for smoking (all/js < .001). Model evaluations using alternative (elliptical and arbitrary distribution) estimators showed the same pattern of significance of structural paths.

Study 5: Illicit Drug Use

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

Adolescent Smoking Time 1

Adolescent Smoking Time 2


Figure 5. Final structural equation model and standardized maximum likelihood estimates for adolescent 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 residuals are depicted with double-headed arrows. Significance levels are based on critical ratios on unstandardized maximum likelihood estimates [*p < .05; **/> < .01; ***p < .001].)

ATTITUDES AND HEALTH BEHAVIOR common variance of marijuana and PCP attitudes, and more general drag-use behavior (a variety of different drugs). However, on the PCP measures only a few respondents indicated values above the minimum. In addition, the preliminary model did not yield any significant paths between the general factors, although it did support the significance of some paths between the marijuana indicators of these factors. Because only marijuana indicators had significant effects, and because the measurement of PCP was problematic, we only report the analysis of a model that was restricted to the associations between marijuana behavior 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 a single indicator. Otherwise, the model was constructed similarly to the models in the previous studies. To maintain compatibility with the other studies, only subjects who had smoked marijuana at least once at the initial time of measurement were used in the analysis (N = 357).


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 from the initial model in terms of model fit. Standardized structural parameter estimates for the final model are reported in Figure 6. Because the model used only single indicators rather than factors, there were no factor loadings. Model evaluations using alternative (elliptical and arbitrary distribution) estimators showed the same pattern of significance of structural paths. In several supplementary analyses, adjustments of the assumed amount of error variance in the single-indicator constructs did not change these results. Discussion Many intervention programs include attitude change as an important element of health-behavior change. In addition, previous research, as well as most of the present studies, has shown significant cross-sectional correlations between attitudes and behavior, suggesting that attitudes may be important precursors of health-related behavior. However, the most general conclusion from the present longitudinal results is that, when stabilities and cross-sectional correlations were taken into account, attitude was neither a consistent nor strong predictor of the health behaviors we assessed. Although attitude did not have general or strong effects, it did predict certain specific health behaviors in two of the six models we analyzed, suggesting that attitude still may be of some importance.

Results The primary path model just described for marijuana use estimated paths from Tl attitude and behavior indicators to T2 attitude and behavior indicators, as well as a covariance between the Tl indicators and a covariance between the T2 residuals of the predicted indicators. Because this initial model was completely saturated (i.e., no additional parameter estimates could be made in the model), it fit the data perfectly (x2 [0, N = 357] = 0). This model was used in model comparisons with more-restricted models, using the procedures reported in the previous studies. One cross-lagged path, from Tl marijuana use to T2 attitude toward marijuana, was found to be nonsignificant; this path was omitted to derive the final model.

Adolescent Marijuana Use Time 1

Adult Marijuana Use Time 2



Attitude Time 1

Attitude Time 2

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


A. STACY, P. BENTLER, AND B. FLAY two-dimensional view of affective predispositions in health behavior. In some health-behavior research, a two-dimensional view has been used to construct separate positive and negative value x expectancy scales representing both belief and affective elements of behavioral tendencies (Stacy, Dent, et al., 1990; Stacy, Widaman, & Marlatt, 1990). Although these constructs have been referred to as expectancies, they could be considered to be quasi-attitudinal constructs because they include affective components. However, instead of merely predicting attitude, in accord with traditional attitude perspectives (e.g., Fishbein & Ajzen, 1975), positive expectancies had a unique and superior predictive effect on behavioral intentions (Stacy, Widaman, & Marlatt, 1990). In addition, positive and negative expectancies equally and strongly predicted attitude, yet positive and negative expectancies were only modestly intercorrelated. Other recent studies, using more traditional assessments of outcome expectancies, also have found that general factors of expectancies about positive and negative outcomes were correlated only weakly, or nonsignificantly (Leigh & Stacy, 1993; Stacy, MacKinnon, & Pentz, 1993), providing further evidence that these constructs are not simply cognitive elements of opposite poles of attitude. Empirical support of the two-dimensional views of affect and expectancies suggests that traditional attitude constructs may need to be complemented, if not replaced, by alternative affective and cognitive constructs in some areas of healthbehavior research. It is possible that attitudes are merely judgments that people sometimes make but that do not represent the underlying source of judgments or behavior. This source may be somewhat less unitary or abstract than attitude and may be better represented by sets of anticipated outcomes or expectancies. If so, then attitude, as traditionally defined and measured in the present studies, would not be expected to be a strong predictor of behavior. Even though this perspective is only one of several different ways to interpret the lack of strong, general effects of attitude on health behavior, the previous success of expectancies as long-term predictors of some health behaviors (Stacy, Newcomb, & Bentler, 1991) suggests further that the expectancy framework may be a viable complement (or alternative) to traditional attitude frameworks. Although alternatives to attitude may be needed to more adequately reflect affective and cognitive influences on health behavior, attitude was predictive of health behavior in two of the models we analyzed (college alcohol use and community marijuana use). Past behavior was a nonspurious predictor of attitude in two other models (adolescent smoking and college binge eating). A third interpretation of the pattern of results is that the nature of the specific behaviors, the population sampled, or the time interval used in these models may have contributed to the obtained differences. In other words, attitudes may be more generally important than we found, but something correlated with the particular analyzed model moderated the predictive effects of attitude on health behavior. Although it is unclear what specific aspects of the populations or the health behaviors could have acted as moderators of attitude-behavior relations in the present studies, we offer some speculations about potential moderators. It is possible

The failure of attitude to be consistently predictive among subjects who have a minimal degree of behavioral experience does not imply that affective or cognitive approaches are not important in health behavior. On the contrary, the lack of direct effects of attitude in most of our analyses suggests that other approaches involving affective or cognitive constructs should be considered as viable alternatives to the direct-effect attitude model we investigated. We now outline several of the possible reasons why attitude may not be a consistent, directeffect predictor of health behavior. First, it is possible that attitude toward health behavior is useful empirically, theoretically, and practically only in the context of a more general theory than that implied by our investigation 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 research on attitudes toward health behavior may find attitude more generally explanatory in the context of one of these formal theories. If so, health-behavior interventions may be improved by an acknowledgment of the complexity of attitude effects on behavior as well as by the explicit use of propositions from these theories in programs of change. Although some healthbehavior interventions use empirically supported, theoretically driven programs of attitude-behavior change, many others do not go beyond a simple direct-effect model of attitude effects. There are other available interpretations. Our study used one of the most typical definitions of attitude, which was represented as a single bipolar affective predisposition toward performing a behavior. In this approach, a person cannot hold strong positive and negative attitudes simultaneously without responding to the attitude measure in the middle range of the single attitude continuum. This approach assumes that positive and negative affect toward a behavior converge or balance out in a type of summary of affective tendencies, which then influences behavior (or a mediator of behavior, e.g., intention; Fishbein & Ajzen, 1975). One example of such an approach regarding health behavior has been outlined by Akers (1992), in which the stronger affective tendencies win out in the overall attitudinal predisposition toward a behavior. Obviously, people can answer attitude questions in this way, because the bipolar construction of attitude scales is likely to force some consideration of both positive and negative affect when answering the questions. However, this does not imply necessarily that attitude is a unitary entity that is represented in this way in the mind. In some perspectives, affect is not a unitary summary of positive and negative qualities of a behavior or object, but is instead two separate dimensions of positive and negative affect. These perspectives are in agreement with findings showing that affect often appears as two dimensions of positive and negative (e.g., Diener & Emmons, 1984; Gotlib & Meyer, 1986; Watson, 1988) and with evidence that each of these dimensions has different neurophysiological substrates (Baker, Morse, & Sherman, 1987; Wise, 1988). The rwo-dimensionai view of affect can be translated into a



that alcohol and marijuana use in mostly nonproblematic populations (college and community populations) are under greater volitional control; for most people sampled from these populations, these behaviors probably are not particularly addictive or compulsive. On the other hand, the literature on binge eating and smoking suggests that these behaviors are often quite compulsive or addictive in quality, even among otherwise nonproblematic populations (e.g., college or high school students). Alcohol use in groups of drunk-driving offenders is by definition more problematic, and is likely to be more compulsive or addictive than alcohol use among college students. This framework would suggest that, among more problematic samples or behaviors, previous behavior (as an indicator of habit) would be a very strong predictor of future behavior, overriding any potential effects of attitudes. In fact, in the present models, past behavior was a stronger predictor of future behavior in the four instances in which attitude was not predictive, compared with the two instances in which attitude was predictive of health behavior. It also is possible that when behavior is particularly stable over time, it has a greater potential to influence attitudes, as found in our binge eating and smoking models. This effect could occur through stable and repeated self-perceptions (Bern, 1978) or other processes (Stacy et al., 1991). Another feature of the models we analyzed that may moderate the predictive effects of attitude is the time interval used to evaluate attitude-behavior relations. In several theories, such as self-perception, cognitive dissonance, and attitude accessibility, the influence of attitudes and behavior on one another may be quite temporary and may not be captured by the moderate to extremely long time intervals assessed in our studies. However, there was no apparent correlation between strength of attitude-behavior relations in our studies and time interval. In fact, one of the only models to show a significant prediction of health behavior by attitudes was in the longest study, in which marijuana attitudes predicted marijuana use over a 9-year interval. Nevertheless, it is possible that much shorter time intervals (e.g., days instead of weeks, months, or years) would show a quite different pattern of findings. Although findings from very short interval studies would be of great theoretical interest, their applied importance is not readily apparent. Most attitude-change interventions in health behavior seek to make long-term changes in behavior. Further research may delineate more specifically what features of the population and specific behavior are responsible for differences in attitude-behavior relations in health. The potential moderators mentioned earlier should be considered for use in future research on attitudes in health behavior. The investigation of such moderators may reveal that attitudes have a relevant degree of predictive utility under certain sets of specific conditions. If so, then developers of prevention programs may be better able to determine the conditions under which an attitude-change program may be most successful. On the other hand, future research may confirm the alternative view that attitudes are too limited in their predictive effects to be useful in health-behavior interventions. Such findings would provide further motivation for the use of alternative affective and cognitive constructs.


Ajzen, I., & Fishbein, M. (1977). Attitude-behavior relations: A theoretical analysis and review of empirical research. Psychological Bulletin, 84, 888-918. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice Hall. Ajzen, I., & Madden, T. J. (1986). Prediction of goal-directed behavior: 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 a general theory. American Sociological Review, 44, 636-655. American Psychiatric Association. (1987). Diagnostic and statistical manual of mental disorders (3rd ed., rev.). Washington, DC: Author. Baker, T. B., Morse, E., & Sherman, J. E. (1987). The motivation to use 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 objective measure 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 the Ben-Tovim Walker Body Attitude Questionnaire (BAQ), a new measure of women's attitudes towards their own bodies. Psychological Medicine, 21, 775-784. Bentler, P. M. (1989). EQS structural equations program manual. Los Angeles: BMDP Statistical Software, Inc. Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness of fit in the analysis of covariance structures. Psychological Bulletin, 88, 588-606. Bentler, P. M., & Speckart, G. (1979). Models of attitude-behavior relations. Psychological Review, 86, 452^64. Bentler, P. M., & Speckart, G. (1981). Attitudes "cause" behaviors: A structural equation analysis. Journal of'Personality and Social Psychology, 40, 226-238. Biglan, A., Gallison, C., Ary, D., & Thompson, R. (1985). Expired air carbon monoxide and saliva thiocyanate: Relationships to selfreports of marijuana and cigarette smoking. Addictive Behaviors, 10, 137-144. Bollen, K. A. (1989). Structural equations with latent variables. New York: Wiley. Brunswick, A. F., & Messeri, P. (1984). Causal factors in onset of adolescents' cigarette smoking: A prospective study of urban Black youth. Advances in Alcohol and Substance Abuse, 3, 35-52. Bruvold, W. H. (1990). A meta-analysis of the California school-based risk reduction program. Journal of Drug Education, 20, 139-152. Charlton, A., & Blair, V. (1989). Predicting the onset of smoking in boys and girls. Social Science and Medicine, 29, 813-818. Corey, S. M. (1937). Professed attitudes and actual behavior. The Journal of Educational Psychology, 28, 271-280. Diener, E., & Emmons, R. A. (1984). The independence of positive and negative affect. Journal of Personality and Social Psychology, 47, 1105-1117. Evans, R. I., Hansen, W. B., & Mittelmark, M. B. (1977). Increasing the validity of self-reports of smoking behavior in children. Journal of Applied Psychology, 62, 521-523. Farquhar, J. W. (1991). The Stanford cardiovascular disease prevention programs. Annals of the New York Academy of Sciences, 623, 327-331.



Murray, D. M., & Perry, C. L. (1987). The measurement of substance use among adolescents. When is the "bogus pipeline" method needed? Addictive Behaviors, 12, 225-233. National Cancer Institute. (1986). Smoking, tobacco, and cancer program: 1985 report (NIH publication No. 86-2687). Washington, DC: National Institute of Health. Newcomb, M. D., & Bentler, P. M. (1988). Consequences of adolescent drug 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 adolescent smoking behavior: Rationale and methods. Journal of Behavioral Medicine, 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 trial for primary prevention of adolescent drug abuse: Effects on drug use prevalence. Journal of the American Medical Association, 261, 32593266. Polivy, J., & Herman, C. P. (1985). Dieting and binging: A causal analysis. 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, treatment, and research (pp. 104-122). New York: Springer. Raden, D. (1985). Strength-related attitude dimensions. Social Psychological Quarterly, 48, 312-330. Rindskopf, D. (1984). Structural equation models: Empirical identification, Heywood cases, and related problems. Sociological Methods & Research, 13, 109-119. Rogosa, D. (1980). A critique of cross-lagged correlation. Psychological Bulletin, 88, 245-258. Rosenberg, M. J., & Hovland, C. I., (1960). Cognitive, affective, and behavioral components of attitudes. In C. I. Hovland & M. J. Rosenberg (Eds.), Attitude organization and change (pp. 1-14). New Haven, CT: Yale University Press. Schlegel, R. P., Manske, S. R., & d'Avernas, J. R. (1985). Alcohol and drug 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 structures under elliptical distributions. Journal of the American Statistical Association, 82, 1092-1097. Snow, W. H., Gilchrist, L. D., & Schinke, S. P. (1985). A critique of progress in adolescent smoking prevention. Children and Youth Services Review, 7, 1-19. Snyder, M. (1974). The self-monitoring of expressive behavior. Journal of Personality and Social Psychology, 30, 526-537. Speckart, G., & Bentler, P. M. (1982). Application of attitudebehavior models to varied content domains. Academic Psychology Bulletin, 4, 453-465. Spencer, J. A., & Fremouw, W. J. (1979). Binge eating as a function of restraint 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 of outcome expectancies on adolescent smokeless tobacco use. Journal of 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 of smoking among adolescents. Psychological Assessment, 2, 442-446. Stacy, A. W., MacKinnon, D. P., & Pentz, M. A. (1993). Generality and specificity in health behavior: Application to warning label and

Fazio, R. H., & Williams, C. J. (1986). Attitude accessibility as a moderator of the attitude-perception and attitude-behavior relations: An investigation of the 1984 presidential election. Journal of Personality and Social Psychology, 51, 505-514. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior: An introduction to theory and research. Reading, MA: AddisonWesley. Flay, B. R. (1987). Mass media and smoking cessation: A critical review. 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 of preventing it. In P. McGrath & P. Firestone (Eds.), Pediatric and adolescent behavioral medicine (pp. 132-183). New York: SpringerVerlag. 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-psychological smoking prevention programs effective? The Waterloo Study. Journal 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 multiple affect adjective check list: A separation of positive and negative affect. Journal of Personality and Social Psychology, 50, 1161-1165. Hawkins, R. C., II, & Clement, P. F. (1980). Development and construct validation of a self-report measure of binge eating tendencies. Addictive Behaviors, 5, 219-226. Huba, G. J., & Harlow, L. L. (1987). Robust structural equation models: Implications for developmental psychology. Child Development, 58, 147-166. Jarvis, M. J., Goddard, E., & McNeil, A. (1990). Do attitudes predict uptake of smoking in teenagers? Case not proven. Social Science and Medicine, 31, 997-1000. Johnson, V. (1988). Adolescent alcohol and marijuana use: A longitudinal assessment of a social learning perspective. American Journal of Drug and Alcohol Abuse, 14, 419-439. Jones, E. E., & Sigall, H. (1971). The bogus pipeline: A new paradigm for measuring affect and attitude. Psychological Bulletin, 76, 349-364. Joreskog, K. G. (1979). Statistical estimation of structural models in longitudinal-developmental investigations. In J. R. Nesselroade & P. B. Baltes (Eds.), Longitudinal research in the study of behavior and development (pp. 303-351). San Diego, CA: Academic Press. Kahle, L. R., & Herman, J. J. (1979). Attitudes cause behavior: A cross-lagged panel analysis. Journal of Personality and Social Psychology, 37, 315-321. Katzman, M. A., & Wolchik, S. A. (1984). Bulimia and binge eating in college women: A comparison of personality and behavioral characteristics. Journal of Consulting and Clinical Psychology, 52, 423-428. Kelman, H. C. (1974). Attitudes are alive and well and gainfully employed in the sphere of action. American Psychologist, 29, 310324. 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 confirmatory models. Psychological Assessment, 5, 216-229. Leventhal, H., & Cleary, P. D. (1980). The smoking problem: A review of research and theory in behavioral risk modification. Psychological Bulletin, 88, 370-408. Madden, T. J., Ellen, P. S., & Ajzen, I. (1992). A comparison of the theory 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-fit indices in confirmatory factor analysis: The effect of sample size. Psychological Bulletin, 103, 391-410.

ATTITUDES AND HEALTH BEHAVIOR social influence expectancies. Journal of Applied Psychology, 78, 611-627. Stacy, A. W., Newcomb, M. D., & Bentler, P. M. (1991). Cognitive motivation 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 of dieting and binge eating. Paper presented at the annual meeting of the 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 multitraitmultimethod assessment. Journal of Personality and Social Psychology, 49, 219-232. Stacy, A. W., Widaman, K. F., & Marlatt, G. A. (1990). Expectancy models of alcohol use. Journal of Personality and Social Psychology, 58, 918-928.


Thurstone, L. L. (1931). The measurement of social attitudes. Journal of Abnormal and Social Psychology, 26, 249-269. Watson, D. (1988). The vicissitudes of mood measurement: Effects of varying descriptors, time frames, and response formats on measures of positive and negative affect. Journal of Personality and Social Psychology, 55, 128-141. Wicker, A. W. (1971). An examination of the "other variables" explanation of attitude-behavior inconsistency. Journal of Personality and Social Psychology, 19, 18-30. Wise, R. A. (1988). The neurobiology of craving: Implications for the understanding and treatment of addiction. Journal of Abnormal Psychology, 97, 118-132. Wyer, R. S., Jr., & Srull, T. K. (1989). Memory and cognition in its social context. Hillsdale, NJ: Erlbaum.


Today's Date:_

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




DATE YOUR ORDER WAS MAILED (OR PHONED) _CHECK ___CHARGE CHECIWCARD CLEARED DATE:_ STATE/COUNTRY (If possible, send a copy, front and back, of your cancelled check to help us in our research of your claim.) ISSUES: MISSING DAMAGED





Thank you. Once a claim is received and resolved, delivery of replacement issues routinely takes 4-6 weeks. ------------ DATE RECEIVED:. ACTION TAKEN: _ STAFF NAME: (TO BE FILLED OUT BY APA STAFF) ^---- 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.


13 pages

Report File (DMCA)

Our content is added by our users. We aim to remove reported files within 1 working day. Please use this link to notify us:

Report this file as copyright or inappropriate


Notice: fwrite(): send of 205 bytes failed with errno=32 Broken pipe in /home/ on line 531