A Structural Equation Model of Social Influences ... - Semantic Scholar

BASIC AND APPLIED SOCIAL PSYCHOLOGY, 28(2), 117–129 Copyright © 2006, Lawrence Erlbaum Associates, Inc.

MODEL ET TICKLE OF AL. ADOLESCENT SMOKING

A Structural Equation Model of Social Influences and Exposure to Media Smoking on Adolescent Smoking Jennifer J. Tickle and Jay G. Hull Dartmouth College

James D. Sargent and Madeline A. Dalton Dartmouth Medical School

Todd F. Heatherton Dartmouth College

This research examined the role of movie portrayals of smoking as well as peer, parent, and sibling smoking as predictors of intentions to smoke and smoking behavior in adolescents. Structural equation modeling was used to assess the fit of a model that proposes that identification with smokers, normative beliefs about smoking, and positive expectations about smoking are mediators of the association between social influences and smoking intentions. Our models provided a good fit for both a cross-sectional model of smoking and a model of smoking initiation. The relationship between media exposure to smoking and intentions to smoke was mediated by positive expectancies and identification as a smoker in the cross-sectional model, and by positive expectancies in baseline neversmokers. Our results indicate that viewing smoking in movies is an important predictor of smoking among adolescents and that identity processes and expectancies serve as mediators of this effect.

Each day approximately 4,800 adolescents try their first cigarette, and more than 2,000 become established smokers (Gilpin, Choi, Berry, & Pierce, 1999). High initiation rates coupled with increasing evidence of the negative health consequences of tobacco use make it clear that smoking by America’s youth is a serious societal issue. This research examined how an important social influence factor, observing smoking in movies, leads to smoking initiation among adolescents and investigates several mediators of that relationship. During identity development, teenagers negotiate old and new self-views and experiment with behaviors that they associate with being cool, independent, and well liked (Marcia, 1983). These identity strivings leave them vulnerable to social influences, especially family, peers, and media stars. In Correspondence should be addressed to Jennifer Tickle, Department of Psychology, St. Mary’s College of Maryland, St. Mary’s City, MD 20686. E-mail: [email protected]

line with social cognitive theory (Bandura, 1986), these role models enact behaviors that adolescents observe and imitate. Researchers have identified a number of social influence variables that contribute to the initiation of smoking, including parent and sibling smoking (e.g., Greenlund, Johnson, Webber, & Berenson, 1997; Jackson, Henriksen, Dickinson, Messer, & Robertson, 1998) and peer smoking (e.g., Chassin, Presson, Sherman, Corty, & Olshavsky, 1984; Collins et al., 1987; Eckhardt, Woodruff, & Elder, 1994; Friedman, Lichtenstein, & Biglan, 1985). In addition to the influence of family and peers, recent research has found that movies may also provide models of smoking that influence adolescents to try tobacco. Smoking in film is prevalent: 87% of top box office films contain tobacco use (Sargent, Tickle, et al., 2001), and at least a quarter of major characters in film smoke (Dalton, Tickle, et al., 2002; Stockwell & Glantz, 1997). Most smokers in film are celebrities, increasing their influence as endorsers of smoking. Sargent, Beach, and colleagues (2001) found that the

118

TICKLE ET AL.

likelihood of adolescents’ trying smoking increased as their exposure to smoking in movies increased. This relationship remained significant after adding other risk factors to the model including sociodemographic characteristics, family and peer smoking, personality characteristics of the child, school performance, and parenting. In a follow-up study, Dalton and colleagues (2003) found that exposure to smoking in movies at baseline in adolescent neversmokers predicted initiation of smoking 1 to 2 years later. Studies have also examined the relationship of smoking by specific stars to adolescent smoking. Distefan, Gilpin, Sargent, and Pierce (1999) found that adolescent smokers were more likely than adolescent neversmokers to prefer stars who smoked, either on or off screen. In addition, adolescents who had experimented with cigarettes and who preferred stars that the regular smokers had selected were nearly 1.5 times more likely to be further along in the smoking initiation process. Tickle, Sargent, Dalton, Beach, and Heatherton (2001) found that adolescents whose favorite star smoked in films were 3 times more likely to have a higher smoking status than adolescents whose favorite star did not smoke in films. Furthermore, adolescents who had never smoked a cigarette but whose favorite star smoked were 16 times more likely to have positive attitudes about smoking than those who selected a nonsmoking star. In a longitudinal study, female adolescents with favorite stars who smoked were 2 times more likely to have tried smoking 3 years later than respondents whose favorite stars did not smoke (Distefan, Pierce, & Gilpin, 2004). Laboratory studies have also found that viewing stars smoking in film increased self-ratings of the likelihood of smoking (Hines, Saris, & ThrockmortonBelzer, 2000) and led to more favorable attitudes toward smokers and smoking (Gibson & Maurer, 2000).

THE THEORETICAL MODEL Given past research that links parent, sibling, peer, and media smoking with adolescent behavior, the current research proposed and tested a model that examined variables that may mediate between exposure to these social influences and adolescent smoking. This is the first study to include media exposure to smoking as part of a testable theoretical model that borrows from social and developmental theories of attitude formation, modeling, and social cognition. According to the theory of reasoned action (TRA), attitudes toward a behavior and subjective norms about the behavior contribute to intentions that in turn are causally related to actual behavior (Azjen & Fishbein, 1980). Conrad, Flay, and Hill (1992) provided a review of research linking smoking intentions to smoking behavior. The present theoretical model examined three sets of predictors (identification as a smoker, normative beliefs about smoking, and positive expectancies about smoking) that may mediate the relationship between social influences and smoking intentions.

Included in the first tier of the model were social influences: movie portrayals of smoking, our variable of primary interest, as well as parent, sibling, and peer smoking. We predicted that exposure to these social influences (a) supplies an image of smoking that appeals to adolescent identity, increasing the likelihood of the teenager identifying with others who smoke; (b) creates misperceptions about the normative nature of smoking, increasing estimates of the prevalence of smoking; and (c) fails to make clear the negative consequences of smoking while glamorizing tobacco use, increasing the endorsement of positive expectancies about smoking (similar to attitudes toward the behavior in the TRA). These attitudinal and cognitive changes may then mediate the relationship between exposure and intentions to smoke. Increased Identification With Smokers Adolescents may come to identify with smokers in two ways: They may see themselves as part of a group in which smoking is an accepted behavior, or they may want to become more like a group in which smoking is an accepted behavior (see Burton, Sussman, Hansen, Johnson, & Flay, 1989). Adolescents who see themselves as similar to individuals who smoke may take up smoking because the behavior is consistent with the image they associate with themselves. Consistent with this argument, research has demonstrated that adolescent smokers’ self-concept ratings, as well as ratings by susceptible nonsmokers, are more similar to adolescents’ ratings of the prototypical smoker than the ratings by nonsmokers (Aloise-Young & Hennigan, 1996; Chassin, Presson, Sherman, Corty, & Olshavsky, 1981; Grube, Weir, Getzlaf, & Rokeach, 1984). In a separate vein, adolescents who are unhappy with their image may initiate behaviors to create a new self-image. On-screen behaviors portrayed by movie stars who are perceived as powerful, attractive, wealthy, and glamorous may be particularly salient and desirable to teens looking for a new image. Smoking by people that adolescents admire may cause smoking to be part of their ideal self-image, increasing their likelihood of having intentions to smoke (Aloise-Young & Hennigan, 1996; Barton, Chassin, Presson, & Sherman, 1982; Chassin et al., 1981; Burton et al., 1989). Therefore, we hypothesized that adolescents who are exposed to social influences who smoke, including media stars who tend to glamorize smoking, would be more likely to identify with and see themselves as similar to smokers, which in turn would increase intentions to smoke. Normative Beliefs About Smoking When uncertain about how to behave, people turn to similar others to determine the social norms about that behavior (Festinger, 1954). Norms are in part based on the perception of the prevalence and acceptability of the behavior in the ref-

MODEL OF ADOLESCENT SMOKING

erence group, and exposure to parents, peers, and media stars that smoke may influence smoking norms. Media portrayals in particular misrepresent both the prevalence and characteristics of smokers. Some studies have found that smoking among major characters in movies is higher than smoking rates in the general population and among certain demographic groups (Hazan, Lipton, & Glantz, 1994; McIntosh, Bazzini, Smith, & Wayne, 1998; Stockwell & Glantz, 1997). Misrepresented norms may decrease prohibitions about smoking, decrease the ability to resist peer pressure, and increase intentions to smoke, all of which may lead to adolescent experimentation with smoking. Overestimation of smoking prevalence is a strong predictor of teen smoking (Sussman et al., 1988). Teens who perceive high descriptive norms for smoking are more likely to try smoking, take up regular smoking, and increase their smoking over time (Chassin et al., 1984; Collins et al., 1987). The perception of a norm, regardless of its accuracy, is likely to influence behavior in line with the norm (Miller & Prentice, 1994). College students who had negative attitudes about behaviors such as drinking alcohol and smoking were more likely to engage in those behaviors if they believed that the behaviors were normative (Botvin, Botvin, Baker, Dusenbury, & Goldberg, 1992; Prentice & Miller, 1993). We hypothesized that exposure to smoking by peers, family, and media would be associated with the belief that smoking is prevalent, which in turn would be associated with intentions to smoke. Positive Expectancies About Smoking When adolescents are exposed to positive role models and well-liked peers who smoke, they may develop positive attitudes about smokers and positive expectancies about the consequences of smoking (Blanton, Gibbons, Gerrard, Conger, & Smith, 1997). These positive models are especially effective influences if the portrayals are the only information the teen has about smoking, as is the case with many adolescents whose parents do not smoke. Adolescents whose parents smoke are at higher risk for smoking, so other socialization influences such as media may have less impact on their behavior than on individuals whose media exposure is a primary source of information about smoking (Sargent et al., 2004). Media can lead to positive expectations about the consequences of smoking by portraying smoking characters as being attractive and successful and living glamorous lifestyles. The typical film smoker is a middle-class, white male—a misrepresentation of the type of person who smokes (Everett, Schnuth, & Tribble, 1998; Hazan et al., 1994). Young smokers are often portrayed as sexier, more rebellious, tougher, and more street-smart than their nonsmoking counterparts (Dalton, Tickle, et al., 2002). These portrayals encourage positive associations to smoking: Smoking is enjoyable and relaxing, helps people to be thin,

119

and helps them feel comfortable at parties. The more viewers endorse these types of positive expectancies, the more likely they are to be susceptible to smoking (Dalton, Sargent, Beach, Bernhardt, & Stevens, 1999; Pierce et al., 1993). Therefore, we hypothesized that exposure to social influences that smoke would be associated with increased positive expectancies about smoking, which would be associated with smoking intentions.

Smoking Behaviors and Intentions A variety of outcome measures have been used to measure tobacco use in adolescents, including prebehavioral attitude constructs such as intention, willingness, and susceptibility. In line with the theory of reasoned action (Ajzen & Fishbein, 1980), intentions to smoke are a major predictor of smoking initiation and ongoing smoking among adolescents (Conrad et al., 1992; Eckhardt et al., 1994). Gibbons, Gerrard, Blanton, and Russell (1998) found that both willingness to smoke and intention are important in predicting health behavior. Pierce, Choi, Gilpin, Farkas, & Merritt (1996) combined intentions to smoke with a measure of resistance to peer offers to smoke, and showed this combination variable called susceptibility predicts experimentation up to 4 years later (see also Jackson, 1998; Pierce, Farkas, Evans, & Gilpin, 1995). The current model used several measures of intentions to predict smoking behavior.

MODEL OVERVIEW AND CURRENT RESEARCH The present research used a structural-modeling approach to test a theoretical model (see Figure 1) of smoking uptake among a large sample of early adolescents. Structural equation modeling permits tests of the contribution of individual components of the model as well as the overall utility of the model. The theoretical model suggested that social influences that expose adolescents to images of cigarette smoking would affect intentions to smoke (a primary predictor of smoking initiation) and that these effects would be mediated by identification with smokers, normative beliefs about smoking, and positive expectancies about smoking. The primary social influence factor of interest was exposure to smoking in movies, but peer, parental, and sibling smoking were also examined as part of the full model. Parental restriction of R-rated media was included in the model given its association to exposure to media smoking (Dalton, Ahrens, et al., 2002). The behavioral outcome being modeled was experimentation with tobacco. First, we tested the model using a cross-sectional sample of adolescents that included the full range of smoking outcomes (both neversmokers as well as adolescents who had initiated smoking). Then, we followed a group of baseline neversmokers over 2 years

120

TICKLE ET AL.

FIGURE 1

Theoretical model of social influences and smoking behavior.

to determine the ability of the model to predict smoking initiation.

METHOD: CROSS-SECTIONAL STUDY MODEL OF CONCURRENT SMOKING Procedure and Participants Thirty of 154 middle schools in New Hampshire and Vermont with at least 150 students were randomly selected and asked to participate in a cross-sectional survey. Fifteen schools agreed to participate within 30 days and were enrolled as participating schools. Passive parental consent was obtained for 98% of eligible participants; parents who did not want their child to participate in the study contacted the school to indicate this preference. In September 1999, 93% of eligible participants completed the survey (7% were absent or refused to participate). Proctors administered the voluntary, confidential questionnaire in class to 5,490 adolescents in fifth through eighth grades. These data have been used in previous research (for further details, see Sargent, Beach, et al., 2001), but the present study incorporated measures not examined in previous research and extended that research by using structural equation modeling to examine a model that aims to better explicate the association between media exposure and smoking.

Participants with missing or inconsistent data on variables used in the model1 were omitted from analyses (N = 521, 9.5%). Students with missing data (compared to those retained for analysis) did not differ by gender, but they were more likely to be in fifth grade, to have poorer school performance, to smoke and have friends or family who smoke, and to be receptive to cigarette advertising. The final sample of 4,969 participants was primarily white (89%) and equally distributed by gender. The grade distribution of the sample was as follows: 9% fifth graders, 27% sixth graders, 31% seventh graders, and 33% eighth graders. Measures

Social influence factors. Participants were asked about smoking influences in their immediate environment, including smoking by friends, parents, and siblings. Friend smoking was assessed by asking “How many of your friends smoke cigarettes?” (none, some, most, or all). Nearly two thirds of the sample did not have friends who smoked. For sibling smoking, participants either responded that they did have a sibling who smoked (15%) or that they did not (85%). Parental smoking was measured with two questions: “Does your mother smoke cigarettes?” and “Does your father 1Participants were also excluded for missing data on gender or age because these variables were used for descriptive purposes.


smoke cigarettes?” (yes, no, don’t know, and don’t have a mother/father). Participants who had stepparents were instructed to respond for whomever they spent the most time with. These two questions were combined to create a single parental smoking variable with the following three response categories: neither parent smokes (60%), one parent smokes (24%), or both parents smoke (16%). For each of these variables, higher numbers indicated more exposure to smokers. These three social influences were examined as exogenous variables—they were not caused by any prior construct in the model. Media exposure to smoking was examined as an endogenous latent construct with multiple indicators.2 A content analysis of 601 films from 1988 to 1999 provided data about the amount of smoking in film (see Dalton, Tickle, et al., 2002, for an overview of the method). Tobacco episodes included any time tobacco products appeared on screen and were handled or used by a major or minor character. Smoking content in film was quantified in two primary ways: seconds of on-screen exposure to tobacco use in the film and a count of episodes (or scenes) in which tobacco use appeared during the film. Additional information was coded for each episode, including the actor or actress associated with the smoking and the salience of the smoking for use in calculating specific aspects of exposure to smoking. The results of the content analysis were then linked to the cross-sectional study to estimate movie smoking exposure for each individual in the study. For this purpose, each student was asked to indicate whether he or she had seen each of 50 movies that were randomly selected from the content analysis sample of 601 movies. Each paper survey, therefore, contained identical survey questions except that each survey contained a unique sample of 50 film titles. The sampling of movies for each survey was stratified by movie rating, and the movie titles appeared in random order. On average, participants reported having seen 17 of the 50 films on their lists. The exposure to smoking from each movie viewed by the respondent was summed to create movie smoking-exposure scores for that respondent. These summary measures approximate the “dose” of exposure to smoking in the media and

2We chose to treat media exposure to smoking as an endogenous variable with multiple indicators because we feel that conceptually and statistically this is most appropriate way to model these variables. Teenagers whose parents restrict viewing of R-rated films do see fewer R-rated films and thus, fewer smoking episodes, and it is reasonable that this would be a causal relationship. The multiple indicators of exposure to smoking in media represent different ways of measuring smoking content in media, and the fact that they have high factor loadings on the latent variable is statistical support that they seem to measure the same construct while allowing for unique variance associated with the individual indicators. However, in response to reviewer questions about our treatment of this variable, we also ran a model excluding parental restrictions from the model and treating media exposure to smoking as an exogenous variable that was allowed to correlate with other exogenous variables. Analyzing the model in this way did not change the statistical significance (significant/not) of any paths in the model.

121

were used as indicators of the latent variable in the model representing media exposure to smoking. The following indicator variables were calculated for each participant: the number of tobacco use episodes seen (tobacco exposure), the number of tobacco use episodes seen in R-rated films (R exposure), the number of tobacco use episodes that displayed particularly salient tobacco use (salient uses), and the total seconds of tobacco use exposure in the films (exposure time). As an index of smoking by admired stars, respondents indicated on the survey whether they liked each of 21 popular movie stars. The number of seconds of smoking by any of the liked stars in films released in the 5 years prior to the survey (1995–1999) was used as the star smoking-exposure variable (star smoking).3 To examine receptivity to tobacco advertising, we collected one additional variable that assessed whether the respondent owned, would be willing to use, or both owned and would be willing to use a cigarette promotional item (CPI; for example, T-shirts, backpacks, or hats). A single item measure of parental media restriction was included as an additional exogenous variable given recent research suggesting that adolescents whose parents restrict access to R-rated films are less likely to see R-rated films and are less likely to smoke (Dalton, Ahrens, et al., 2002). The question asked was “How often do your parents let you watch movies or videos that are rated R?” (never, once in a while, sometimes, and all the time). Responses were recoded such that higher numbers indicated more parental restriction. Approximately 16% of the respondents reported that their parents did not ever restrict their movie viewing, whereas 31% reported that their parents restricted R-movies all the time.

Mediators. Three proposed mediators were assessed as endogenous constructs in the model. Identification with smokers was measured with the item “Do you think most kids who are like you start smoking cigarettes?” (Jackson, 1998).4 Normative beliefs were measured with the item “Do you think most adults smoke?” to assess the degree to which smoking was seen as a prevalent behavior (perception of a descriptive norm). Because of the young age of the sample, this prevalence measure was simplified from previous research that required students to estimate the number of indi-

3For estimation of exposure to liked star smoking, the measure assumed that participants see films by stars they like. This measure did not confirm that they had seen the films that contribute to the smoking-exposure estimation. Because this measure examined smoking by liked stars across 21 stars, individuals were not excluded from the analyses if they did not respond to one of the stars. Their scores simply reflect smoking by stars they did report liking. 4Jackson (1998) used this question as a measure of norms related to susceptibility. We used the item as a measure of identification because the question asks specifically about the behavior of “most kids who are like you,” which suggests to us the important role of a comparison group that is similar.

122

TICKLE ET AL.

viduals out of 100 who smoke (e.g., Chassin et al., 1984). Both questions were answered with the options definitely yes, probably yes, probably no, and definitely no. The variables were reverse scored such that higher scores indicated higher identification with smokers and higher normative beliefs about smoking. Positive expectancies were measured using a series of seven items that were each rated on a 4-point scale (strongly agree, agree, disagree, strongly disagree). The items were: “I think I would enjoy smoking,” “I think smoking would give me something to do when I’m bored,” “I think smoking would help me deal with problems or stress,” “I think smoking would help me to be thin,” “I think smoking would help me to feel more comfortable at parties,” “I think smoking would be relaxing,” and “I think smoking would make me look tough.” Strong disagreement was coded as nonendorsement of the expectancy; all other responses were coded as endorsement. The number of items endorsed formed the positive expectancy score, which ranged from 0 (did not endorse any positive expectancies) to 7 (endorsed all of the positive expectancies). This method of coding expectancies is consistent with previous research using these items and has been shown to be a valid predictor of smoking susceptibility (Dalton et al., 1999).

Smoking intentions. Intention to smoke was an endogenous latent construct estimated by six indicator variables.5 Five of these items were rated on a 4-point scale (definitely yes, probably yes, probably no, definitely no): “If one of your friends offered you a cigarette, would you try it?” “Do you think you will smoke a cigarette some time in the next year?” “Would you smoke a cigarette if someone gave you one?” “Do you think you will smoke cigarettes when you are in high school?” “Do you think you will ever smoke cigarettes?” The first two items are validated components of susceptibility to smoking that have been shown to predict adolescent smoking, and the latter three items have also been used as cognitive predictors of smoking (e.g., Jackson, 1998; Pierce et al., 1995; Pierce et al., 1996). The seventh item asked if participants had ever almost tried smoking but then decided not to (yes or no; Jackson, 1998). All intention variables were recoded such that higher scores indicated higher intentions to smoke. Smoking behavior. Level of experimentation with smoking is the behavioral outcome variable. It was assessed by self-report with the question “How many cigarettes have you smoked in your life?” Responses categories were none (indicating a neversmoker; 83% of the sample), just a few puffs (10%), 1–19 cigarettes (3%), 20–100 cigarettes (2%), and over 100 cigarettes (2%). The standard for being consid-

5A seventh item assessed whether the individual wanted to try smoking and had a response option of have already tried smoking. This variable was therefore confounded with the outcome measure of smoking behavior. So, although it is part of the covariance matrix, it was dropped as an indicator of intentions in the structural model.

ered a smoker is having smoked 100 cigarettes in one’s lifetime (Kovar & Poe, 1985). RESULTS: CROSS-SECTIONAL MODEL Modeling was conducted using EQS Structural Equation Modeling Software (Bentler, 1995) and maximum likelihood estimation techniques.6 Model Evaluation Criteria Models were evaluated using multiple fit criteria. One criterion involved the statistical significance of estimated parameters. A second criterion involved fit to the observed data matrix. Lack of a statistically significant χ2 goodness-of-fit test is evidence that the predicted associations among variables are not significantly different from the observed associations. Although we report the χ2 goodness-of-fit test, it is widely acknowledged to have a variety of undesirable properties such as a bias toward significance with large sample sizes (Bentler, 1990; Bollen, 1989; Browne & Cudeck, 1993; Rigdon, 1996). Bentler’s (1990) comparative fit index (CFI) is representative of a class of incremental fit indexes that have achieved wide currency, although it is not associated with a statistical significance test. The CFI is less influenced by sample size and can be interpreted as percent of variance in the covariance matrix accounted for by the model. The root mean square error of approximation (RMSEA or ε) is a fit index for which it is possible to form confidence intervals to reject the hypothesis that a model fits poorly (see Browne & Cudeck, 1993; MacCallum, Browne, & Sugawara, 1996). As with the CFI, the value of RMSEA is a function of the χ2 goodness-of-fit test. This is not true of the final index of fit that we report; the standardized root mean residual (SRMR) is a standardized summary of the average covariance residuals. With respect to the CFI, it has been recommended that models should be evaluated as a good fit if the CFI exceeds .90 (Bentler & Bonett, 1980), although Hu and Bentler (1998, 1999) argued for a more stringent cutoff of .95. In guidelines for the interpretation of RMSEA, MacCallum et al. (1996; see also Browne & Cudeck, 1993) characterized values less than .05 as indicative of “close” fit, values between .05 and .08 as indicative of “fair” fit, values between .08 and .10 as indicative of “mediocre” fit, and values in excess of .10 as indicative of “poor fit.” More recently, Hu and Bentler (1998, 1999) recommended combinations of criteria. Specifically, when the sample exceeds 500, they reported that a value of RMSEA close to .06 together with a value of SRMR close to .10 results in the least likelihood of Type I and Type II error rates in model evaluation.

6The correlation matrices for all measures used in the cross-sectional and initiation models with their means and standard deviations are available by request from the corresponding author.


Given this background, it was our preference to develop models that achieve the Hu and Bentler (1998, 1999) CFI criteria of .95, and we found the MacCallum et al. (1996) guidelines for the interpretation of RMSEA useful as a means of characterizing the general quality of the model’s fit. Specifically, we used the confidence interval property of RMSEA to reject the hypothesis that a model had given a level of poor fit to the observed data. Latent Variable Measurement Model of Media Exposure and Smoking Intentions An initial latent variable model was specified that represented the six measures of media exposure and six measures of smoking intentions in terms of two separate but correlated latent variables. Although this model provided an adequate fit according to some criteria, it showed ill-fit according to other criteria, χ2(53; N = 4,969) = 3,172.54, CFI = .942, SRMR = .085, RMSEA = .109. Specifically, the 90% confidence interval around RMSEA did not allow us to reject the hypothesis of poor fit, .106 ≤ ε ≤ .112. Given that an adequately fitted measurement model was a prerequisite for testing our structural hypotheses, attempts were made to improve the fit of the model. First, modification statistics suggested that the CPI use/ownership measure was more strongly associated with the intentions latent variable than the media latent variable. CPI was therefore dropped as a misspecified indicator. The remaining six indicators were clearly alternate measures of a single underlying construct, exposure to smoking in film. Second, specific variances in the individual indicators were allowed to correlate as long as they were conceptually sensible and involved variables that had already been specified as indicators of the same latent variable.7 These modifications were made so that subsequent ill-fit in the structural model could be attributed to misspecifications of theorized relations between variables of conceptual interest and not to an inadequately specified measurement model. This process resulted in the inclusion of three correlated errors, a substantial improvement in fit, difference χ2(3; N = 4,969) = p < .0001, and an excellent fit of the observed data, χ2(40; N = 4,969) = 307.126, CFI = .995, SRMR = .034, RMSEA = .037. The 90% confidence interval around RMSEA allowed rejection of the hypothesis of anything less than excellent fit, .033 ≤ ε ≤ .041. In the final measurement model, four of the five indicators of the media exposure latent variable had factor loadings exceeding .90 and five of the six indicators of Intentions had factor loadings exceeding .77. This suggests that the indicator variables were measuring the same latent constructs.

7Specifically, the errors that were allowed to correlate were (a) respondents’ willingness to try smoking if someone offered them a cigarette and their willingness to try smoking if offered a cigarette by a friend, (b) respondents’ expectation that they would ever smoke and their expectation that they would smoke in high school, and (c) salient smoking episodes and smoking episodes in R-rated films.

123

Cross-Sectional Model Given a properly fitting measurement model, an initial theoretical model was tested using the cross-sectional data.8 This model provided a reasonable fit of the observed data, χ2(134; N = 4,969) = 3,420.071, CFI = .951, SRMR = .101, RMSEA = .070, with the confidence intervals around RMSEA allowing rejection of an hypothesis of mediocre fit, .068 ≤ ε ≤ .072. Although this model provided a reasonable fit of the observed data, it did not meet the Hu and Bentler (1999) joint criteria of a RMSEA value of .06 and an SRMR value less than .10. Lagrange Multiplier (LM) tests suggested that the single largest improvement in fit would occur if peer smoking was theorized to have a direct effect on smoking intentions. Given that such an association is conceptually sensible (see Conrad et al., 1992, for a review), the initial theoretical model was modified to include this path.9 This modified model provided a substantial improvement over the originally specified theoretical model, difference χ2(1; N = 4,969) = 1,031.428, p < .0001, and provided a good overall fit to the observed data, χ2(133; N = 4,969) = 2,388.643, CFI = .966, SRMR = .074, RMSEA = .058, .056 ≤ ε ≤ .060. This model is depicted in Figure 2. As can be seen in Figure 2, the assumption that identification, norms, and expectancies mediate the relationship between the smoking by social influences (peers, siblings, parents, and media) and intentions to smoke is symbolically represented by 15 paths. Of the 12 paths linking smoking by social influences to identification, norms, and expectancies, only 2 are not significant (parent smoking to identification; sibling smoking to positive expectancies). Of the 3 paths linking identification, norms, and expectancies to intentions to smoke, both identification and expectancies are strong predictors of intentions. In addition, in this modified model, peer smoking has an unmediated impact on intentions to smoke. Finally, intentions and actual smoking behavior are strongly linked. 8To confirm the existence of relationships between the social influence variables (peer, parent, sibling, and media smoking) and the outcome variables (intentions and behavior) that could be mediated, we conducted analyses of the direct effects of these variables (also including the indirect effect of parental restriction through media exposure). Each of the social influence variables with the exception of sibling smoking was significantly associated with intentions to smoke, and each of the social influence variables was significantly associated with smoking behavior. 9Given the controversy surrounding the use of modification indexes to improve model fit, we used parallel analyses of random split halves of the sample to insure that the addition of the path between peer smoking and intentions was not spurious (MacCallum et al., 1992; MacCallum et al., 1996; Wegener & Fabrigar, 2000). In both samples, the LM chi square was the largest of all LM estimates and was highly significant (Group 1: LM χ2(1, N = 2,485) = 516.59, p < .001; Group 2: LM χ2(1, N = 2,484) = 462.71, p < .001). Addition of these paths also led to improvement in the cross-validation index of the Akaike information criteria in both the original sample as well as the randomly split subsamples. Therefore, we are reasonably confident that we have not capitalized on chance and that this is a stable additional path.

124

TICKLE ET AL.

FIGURE 2

Cross-sectional model of social influences and smoking behavior.

DISCUSSION: CROSS-SECTIONAL MODEL In the cross-sectional model, peer smoking, sibling smoking, and media exposure to smoking were each associated with increased identification with smokers, which was associated with higher intentions to smoke. This supports the hypothesis that adolescents who are exposed to people who smoke are more likely to identify with smokers, and identification is related to smoking intentions. Contrary to our hypothesis that parents would also be a source of identification, parental smoking was not associated with identification as a smoker in this sample. Parents may not be a group with whom teenagers identify when it comes to smoking, or our measure of identification may not have been a good measure of general identification processes because it asked specifically about identification with similar peers. All four social influence factors were associated with increased normative beliefs about smoking, as was predicted. However, contrary to predictions, increased norms were not associated with intentions to smoke. It should be noted that the question about norms focused on the percentage of adults who smoke, which might not be as valid a predictor of intentions to smoke in younger adolescents as are estimates of same-age smoking (see Chassin et al., 1984). To provide a broader test of the role of norms as would be predicted by the TRA, prevalence estimates of smoking in more varied popu-

lations (for example, same age peers) could be paired with questions about the reactions of close others to one’s smoking and measures of motivation to comply with the perspective of close others. Future research should incorporate additional measures of normative beliefs to take into consideration this expanded concept of norms. As hypothesized, higher levels of peer smoking, parental smoking, and media exposure to smoking were each associated with increased endorsement of positive expectancies about smoking, which was related to intentions to smoke. On the other hand, sibling smoking was not associated with expectations about smoking. It is unclear why sibling smoking would not contribute to expectancies about smoking, and should be examined further in future research. As predicted, parental restriction of media was associated with decreased exposure to smoking in film, supporting previous research that parents can be effective in reducing media exposure to smoking by prohibiting the viewing of rated-R films. Finally, the independent pathway added to the model from peer smoking to smoking intentions underscores the importance of peers as a particularly valuable source of social comparison with regard to smoking. This research supported the hypothesis that media exposure to smoking was related to intentions to smoke through its association with increased positive expectancies about smoking and increased identification as a smoker. However,


it was important to test this model prospectively to examine how these mediators influence the effects of exposure to movie smoking on smoking initiation. The follow-up component of this study set out to understand the effect of media exposure to smoking on smoking initiation. To most effectively focus on smoking initiation, we resurveyed the neversmoking participants from the cross-sectional study because those participants could have started smoking since the initial survey. We then reexamined our model as applied to smoking initiation 1 to 2 years following the initial assessment.

METHOD: MODEL OF SMOKING INITIATION This study followed the cross-sectional participants who provided contact information and who had never smoked, even a few puffs, at baseline (N = 3,547). These respondents were contacted by telephone 13 to 26 months after the baseline survey, and 2,603 students were successfully contacted (a follow-up rate of 73%). Neversmokers who could not be reached for the follow-up survey did not differ from the follow-up sample by age, sex, grade in school, or exposure to movie smoking. However, they were more likely to have friends, siblings, and parents who smoked, have lower school performance, and were slightly more likely to be susceptible to smoking at baseline (see Dalton et al., 2003, for more information). The modified cross-sectional model was used to model smoking initiation. In this theorized model, all variables were identical to those used in the cross-sectional model except the measure of smoking behavior, which was replaced by smoking behavior at the time of the follow-up assessment. Because all participants in this sample responded that they had never smoked at baseline, those participants whose response changed from baseline had initiated smoking and were considered “triers” at follow-up. Ten percent of the participants tried smoking during the follow-up period, and smoking behavior was categorized for analysis as participants who had never smoked (90%), those who had just a few puffs (8%), those who smoked 1–100 cigarettes (1.8%), and those who smoked more than 100 cigarettes (.2%) during the follow-up. After excluding participants with missing data, the final sample for the model of smoking initiation contained 2,541 students. The distribution of responses for most variables, including grade, were similar to the distributions reported in the cross-sectional model.

RESULTS: MODEL OF SMOKING INITIATION Once again, modeling was conducted using EQS (Bentler, 1995) and maximum likelihood estimation techniques. Given a cross-sectional model that provided a good fit of the

125

data from all participants, the model was examined on a selection of those participants (a) who did not smoke at Time 1 and (b) whose smoking status was reassessed at Time 2. Because these participants reported at Time 1 that they did not smoke, smoking at Time 2 constituted initiation of smoking.10 The model provided an excellent fit to this subsample of data, χ2(133; N = 2,541) = 979.290, CFI = .968, SRMR = .066, RMSEA = .050, .047 ≤ ε ≤ .053. This signifies that the model also provided a reasonable model of smoking initiation for students who had not smoked at baseline. Given our past research on the association of exposure to smoking by liked film stars on the initiation of smoking by adolescents (see Tickle et al., 2001), we examined the modification statistics with particular interest in this variable. The modification statistics did suggest that the model fit could be improved with the addition of a direct path from the specific variance associated with exposure to smoking by liked stars (an indicator of the latent media exposure variable) to smoking at Time 2.11 The model was modified to allow this direct path, and the addition of this path resulted in a significant improvement in fit, difference χ2(1; N = 2,541) = 12.13, p < .0005, suggesting that it uniquely contributes to the initiation of smoking by adolescent nonsmokers. The final model was associated with an excellent fit, χ2(132; N = 2,541) = 967.160, CFI = .968, SRMR = .065, RMSEA = .050, .047 ≤ ε ≤ .053, and is depicted in Figure 3. As shown in Figure 3, of the 12 paths linking smoking by social influences to identification, norms, and expectancies, only 5 are significant in this subsample of participants who had never smoked. Although peer smoking was related to all three mediation variables (identification, norms, and expectancies), sibling smoking was only related to normative beliefs, and parental smoking and media exposure were only related to positive expectancies. As in the cross-sectional model, identification as a smoker and positive expectancies were both related to intentions to smoke. Finally, intentions to smoke were related to subsequent initiation of smoking. Of interest given our previous research, initiation of smoking was also related to smoking by liked stars.

10As with the cross-sectional sample, we tested the direct effects of the four social influence variables on intentions, and separately, behavior (including the indirect effect of parental restriction of media through media exposure). Again, all social influences were directly related to intentions with the exception of sibling smoking, and all social influences were directly related to smoking behavior. 11We again used parallel analyses of random split halves of the sample to insure that the addition of the smoking by liked stars pathway was not spurious. In both samples, the LM chi square was statistically significant (Group 1: LM χ2(1, N = 1,271) = 5.65, p = .017; Group 2: LM χ2(1, N = 1,270) = 6.20, p = .013). Addition of these paths also led to improvement in the cross-validation index of the Akaike information criteria in the original sample as well as the randomly split subsamples.

126

TICKLE ET AL.

FIGURE 3

Model of social influences and smoking initiation in neversmokers.

DISCUSSION: MODEL OF SMOKING INITIATION Unlike the cross-sectional model, the model of smoking initiation included only participants who were neversmokers at baseline. Therefore, this model provided more direct evidence of the factors that determine smoking for the first time and had an advantage over the cross-sectional model in that it ruled out the possibility that smoking behavior might have caused increased viewing of smoking in media because all participants were neversmokers at baseline. Having peers who smoke was the only social influence factor related to identification as a smoker, which was in turn related to smoking intentions. In this sample of participants who had never smoked at baseline, siblings and media did not retain their relations to greater identification as a smoker, suggesting that this pathway may be more important after an adolescent has first tried smoking. Similar to the cross-sectional model, having peers and siblings who smoke was related to higher normative beliefs about smoking, but again norms were not associated with smoking intentions. Exposure to peer, parent, and media smoking were associated with endorsement of positive expectancies about smoking, which was in turn related to smoking intentions. In this initiation model, the role of media exposure to smoking seemed to have the strongest association with intentions to smoke through the expectancies vari-

able. This suggests that the positive consequences of smoking that are shown in movies and the positive associations that teens have to the stars who smoke in movies may be an important influence on how they think about the consequences of smoking. One of the most striking effects in this model was that liking stars that smoked frequently in film at the baseline survey was strongly related to having tried smoking up to 2 years later. This relation was not fully accounted for by the associations between media exposure, the mediators, and intentions to smoke, indicating that the influence of stars may be accounting for unique variance in smoking initiation. It is possible that smoking by liked stars is an aspect of media exposure that is conceptually consistent with our idea of identification: Stars who smoke on screen portray attributes that become part of a teen’s ideal self-image. Reconceptualizing star smoking as a distinct mediator rather than an indicator of the media exposure latent construct might provide additional insight into the role of this variable in the larger model of social influences on smoking intention. This model confirmed the negative relationship between parental media restriction and exposure to smoking in media, in that children whose parents restricted media were less likely to begin smoking. In addition, the model retained the added pathway between peer smoking and intentions to smoke, highlighting the importance of peers in the initiation of adolescent smoking.


GENERAL DISCUSSION Given the immense amount of media to which children are exposed, it is imperative to understand the impact of this social influence on smoking and other forms of substance abuse. Our previous research has shown that media exposure to smoking is related to adolescent smoking initiation, even after controlling for sociodemographic variables and the possibility of unmeasured covariates (Dalton et al., 2003). The goal of the present research was to extend this previous research to examine possible mediators of this effect and to support the hypothesis that social influences, including exposure to smoking in movies, were associated with increased identification with smokers and increased positive expectancies about smoking, which were in turn related to intentions to smoke. Although models have examined factors that relate to general adolescent substance use (e.g., McAlister, Krosnick, & Milburn, 1984; Stacy, Sussman, Dent, Burton, & Flay, 1992; Wills & Cleary, 1996), this model is the first to specifically examine media exposure to smoking as part of the theoretical framework and to test mediators that might help explain the influence of media on intentions to smoke. In our model, identification with smokers, normative beliefs about smoking, and positive expectations about smoking were examined as mediators of the effects of peer smoking, family smoking, and media exposure to smoking on smoking intentions and behavior. The model provided good fit to both cross-sectional smoking and smoking initiation. Exposure to smoking in media influenced adolescents to smoke by increasing their positive expectancies about smoking, which in turn led to greater intentions to smoke. Identification as a smoker also mediated the influence of media exposure on intentions to smoke in our cross-sectional sample. Although exposure to peer smoking is the most robust influence in the models, the contribution of media exposure to smoking, particularly smoking by liked stars, remains an important contributor to the process of adolescent smoking. The strength of the paths in the model varied from relatively strong (as with the relationship between peer smoking and positive expectancies and between positive expectancies and intentions) to relatively weak but still statistically significant (as with the relationship between media exposure to smoking and identification in the cross-sectional model). Although the relationship between media exposure and smoking is not as strong as some of the others observed, the public health significance of this association is still quite substantial because of the widespread exposure of adolescents to movie smoking. Whereas only one third of the adolescents in the cross-sectional sample had at least one friend who smoked, many of them had exposure to smoking through media. The strength of the relationship between media exposure and smoking initiation in combination with the prevalence of the exposure makes media a potent influence (Dalton et al.,

127

2003). Given the importance of understanding how exposure to smoking influences adolescent smoking, we feel that all paths that were statistically significant warrant further consideration. We also reiterate the need for more comprehensive measures of identification and normative beliefs to more fully understand the link between media exposure to smoking and smoking initiation. As states continue to ban smoking in public places, adolescents whose families and friends do not smoke may have media as their primary form of exposure to smoking. Given the positive way that smoking is portrayed in film paired with recent research that suggests that media is most influential when other models of smoking are not available (Sargent et al., 2004), it becomes increasingly important to better understand how smoking in media is having its effect on smoking intentions. This research takes an important first step in outlining two possible mechanisms, identification processes and expectancies, that can be addressed in ongoing attempts to find ways to decrease the influence of media smoking on impressionable youth. Future research should continue to explore mediators of the influence of media on adolescent smoking behavior using additional measures and diverse adolescent populations. Nearly a half million Americans die of smoking-related deaths every year, and not only do many individuals first try smoking during adolescence but most addicted smokers also begin smoking in their teens (U.S. Department of Health and Human Services, 1994). In terms of media exposure to smoking, research and policy makers should explore ways to reduce exposure to smoking by enabling parents to restrict media access, by limiting the smoking in movies, by taking the lead of other industrialized nations to teach adolescents to become resistant to movie images through media literacy training, or by examining ways to construct images of smoking that do not communicate positive expectancies or encourage identification with smoking characters. Focusing prevention efforts and policy on reducing exposure to media as well as other social influences that smoke or limiting their impact on adolescent attitudes and intentions to smoke should help reduce the number of adolescents who become addicted adult smokers.

ACKNOWLEDGMENTS This research was supported in part by grants from the National Cancer Institute. Jennifer Tickle is now at the Department of Psychology, St. Mary’s College of Maryland.

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

128

TICKLE ET AL.

Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on Automatic Control AC, 19, 716–723. Aloise-Young, P. A., & Hennigan, K. M. (1996). Self-image, the smoker stereotype and cigarette smoking: Developmental patterns from fifth through eighth grade. Journal of Adolescence, 19, 163–177. Bandura, A. (1986). Social foundations of thought and action. Englewood Cliffs, NJ: Prentice Hall. Barton, J., Chassin, L., Presson, C. C., & Sherman, S. J. (1982). Social image factors as motivators of smoking initiation in early and middle adolescence. Child Development, 53, 1449–1511. Bentler, P. M. (1990). Comparative fit indexes in structural models. Psychological Bulletin, 107, 238–246. Bentler, P. M. (1995). EQS: Structural equations program manual. Encino, CA: Multivariate Software. Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness of fit in the analysis of covariance structures. Psychological Bulletin, 88, 588–606. Blanton, H., Gibbons, F. X., Gerrard, M., Conger, K. J., & Smith, G. E. (1997). Role of family and peers in the development of prototypes associated with substance use. Journal of Family Psychology, 11, 271–288. Bollen, K. A. (1989). Structural equations with latent variables. Oxford, England: Wiley. Botvin, G. J., Botvin, E. M., Baker, E., Dusenbury, L., & Goldberg, C. J. (1992). The false consensus effect: Predicting adolescents’ tobacco use from normative expectations. Psychological Reports, 70, 171–178. Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K. A. Bollen & J. S. Long (Eds.), Testing structural equation models (pp. 136–162). Thousand Oaks, CA: Sage. Burton, D., Sussman, S., Hansen, W. B., Johnson, C. A., & Flay, B. R. (1989). Image attributions and smoking intentions among seventh grade students. Journal of Applied Social Psychology, 19, 656–664. Chassin, L., Presson, C. C., Sherman, S. J., Corty, E., & Olshavsky, R. O. (1981). Self-images and cigarette smoking in adolescents. Personality and Social Psychology Bulletin, 7, 670–676. Chassin, L., Presson, C. C., Sherman, S. J., Corty, E., & Olshavsky, R. O. (1984). Predicting the onset of cigarette smoking in adolescents: A longitudinal study. Journal of Applied Social Psychology, 14, 224–243. Collins, L. M., Sussman, S., Mestel-Rauch, J., Dent, C. W., Johnson, C. A., Hansen, W. B., et al. (1987). Psychosocial predictors of young adolescent cigarette smoking: A sixteen-month, three-wave longitudinal study. Journal of Applied Social Psychology, 17, 554–573. Conrad, K. M., Flay, B. R., & Hill, D. (1992). Why children start smoking cigarettes: Predictors of onset. British Journal of Addiction, 87, 1711–1724. Dalton, M. A., Ahrens, M. B., Sargent, J. D., Mott, L. A., Beach, M. L., Tickle, J. J., et al. (2002). Relation between adolescent use of tobacco and alcohol and parental restrictions on movies. Effective Clinical Practice, 5, 1–10. Dalton, M. A., Sargent, J. D., Beach, M. L., Bernhardt, A. M., & Stevens, M. (1999). Positive and negative outcome expectations of smoking: Implications for prevention. Preventive Medicine, 29, 460–465. Dalton, M. A., Sargent, J. D., Beach, M. L., Titus-Ernstoff, L., Gibson, J. J., Ahrens, M. B., et al. (2003). Effect of viewing smoking in movies on adolescent smoking initiation: A cohort study. The Lancet, 357, 29–32. Dalton, M. A., Tickle, J. J., Sargent, J. D., Beach, M. L., Ahrens, M. B., & Heatherton, T. F. (2002). The incidence and context of tobacco in popular movies from 1988–1997. Preventive Medicine, 34, 516–523. Distefan, J. M., Gilpin, E. A., Sargent, J. D., & Pierce, J. P. (1999). Do movie stars encourage adolescents to start smoking? Evidence from California. Preventive Medicine, 28, 1–11. Distefan, J. M., Pierce, J. P., & Gilpin, E. A. (2004). Do favorite movie stars influence adolescent smoking initiation? American Journal of Public Health, 94, 1239–1244. Eckhardt, L., Woodruff, S. I., & Elder, J. P. (1994). A longitudinal analysis of adolescent smoking and its correlates. Journal of School Health, 64, 67–72.

Everett, S. A., Schnuth, R. L., & Tribble, J. L. (1998). Tobacco and alcohol use in top-grossing American films. Journal of Community Health, 23, 317–324. Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140. Friedman, L., Lichtenstein, E., & Biglan, A. (1985). Smoking onset among teens: An empirical analysis of initial situations. Addictive Behaviors, 10, 1–13. Gibbons, F. X., Gerrard, M., Blanton, H., & Russell, D. W. (1998). Reasoned action and social reaction: Willingness and intention as independent predictors of health risk. Journal of Personality and Social Psychology, 74, 1164–1181. Gibson, B., & Maurer, J. (2000). Cigarette smoking in the movies: The influence of product placement on attitudes toward smoking and smokers. Journal of Applied Social Psychology, 30, 1457–1473. Gilpin, E. A., Choi, W. S., Berry, C., & Pierce, J. P. (1999). How many adolescents start smoking each day in the United States? Journal of Adolescent Health, 25, 248–255. Greenlund, K. J., Johnson, C. C., Webber, L. S., & Berenson, G. S. (1997). Cigarette smoking attitudes and first use among third- through sixth-grade students: The Bogalusa Heart Study. American Journal of Public Health, 87, 1345–1348. Grube, J. W., Weir, I. L., Getzlaf, S., & Rokeach, M. (1984). Own value system, value images, and cigarette smoking. Personality and Social Psychology Bulletin, 10, 306–313. Hazan, A. R., Lipton, H. L., & Glantz, S. A. (1994). Popular films do not reflect current tobacco use. American Journal of Public Health, 84, 998–1000. Hines, D., Saris, R. N., & Throckmorton-Belzer, L. (2000). Cigarette smoking in popular films: Does it increase viewers’ likelihood to smoke? Journal of Applied Social Psychology, 30, 2246–2269. Hu, L., & Bentler, P. M. (1998). Fit indices in covariance structure modeling: Sensitivity to underparameterized model misspecification. Psychological Methods, 3, 424–453. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 16, 1–55. Jackson, C. (1998). Cognitive susceptibility to smoking and initiation of smoking during childhood: A longitudinal study. Preventive Medicine, 27, 129–124. Jackson, C., Henriksen, L., Dickinson, D., Messer, L., & Robertson, S. B. (1998). A longitudinal study predicting patterns of cigarette smoking in late childhood. Health Education and Behavior, 25, 436–447. Kovar, M. G., & Poe, G. S. (1985). The National Health Interview Survey design, 1973–84, and procedures, 1975–83. Washington DC: U.S. Department of Health and Human Services, Public Health Service, National Center for Health Statistics. MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and determination of sample size for covariance structure modeling. Psychological Methods, 1, 130–149. MacCallum, R. C., Roznowski, M., & Necowitz, L. B. (1992). Model modifications in covariance structure analysis: The problem of capitalization on chance. Psychological Bulletin, 111, 490–504. Marcia, J. E. (1983). Some directions for the investigation of ego development in early adolescence. Journal of Early Adolescence, 3, 215–223. McAlister, A. L., Krosnick, J. A., & Milburn, M. A. (1984). Causes of adolescent cigarette smoking: Tests of a structural equation model. Social Psychology Quarterly, 47, 24–36. McIntosh, W. D., Bazzini, D. G., Smith, S. M., & Wayne, S. M. (1998). Who smokes in Hollywood? Characteristics of smokers in popular films from 1940 to 1989. Addictive Behavior, 23, 395–398. Miller, D. T., & Prentice, D. A. (1994). Collective errors and errors about the collective. Personality and Social Psychology Bulletin, 20, 541–550. Pierce, J. P., Choi, W. S., Gilpin, E. A., Farkas, A. J., & Merritt, R. K. (1996). Validation of susceptibility as a predictor of which adolescents take up smoking in the United States. Health Psychology, 15, 355–361.

MODEL OF ADOLESCENT SMOKING Pierce, J. P., Farkas, A., Evans, N., Berry, C., Choi, W., Rosbrook, B., et al. (1993). Tobacco use in California 1992. A focus on preventing uptake in adolescents. Sacramento: California Department of Health Services. Pierce, J. P., Farkas, A. J., Evans, N., & Gilpin, E. (1995). An improved surveillance measure for adolescent smoking? Tobacco Control, 4, 47–56. Prentice, D. A., & Miller, D. T. (1993). Pluralistic ignorance and alcohol use on campus: Some consequences of misperceiving the social norm. Journal of Personality and Social Psychology, 64, 243–256. Rigdon, E. E. (1996). CFI vs. RMSEA: A comparison of two fit indices for structural equation modeling. Structural Equation Modeling, 3, 369–379. Sargent, J. D., Beach, M. L., Dalton, M. A., Mott, L. A., Tickle, J. J., Ahrens, M. B., et al. (2001). Effect of seeing tobacco use in films on trying smoking among adolescents: Cross-sectional study. British Medical Journal, 323, 1394–1397. Sargent, J. D., Beach, M. L., Dalton, M. A., Titus Ernstoff, L., Gibson, J. J., Tickle, J. J., et al. (2004). Effect of parental R-rated movie restriction on adolescent smoking initiation: A prospective study. Pediatrics, 114, 149–156. Sargent, J. D., Tickle, J. J., Beach, M. L., Dalton, M. A., Ahrens, M. B., & Heatherton, T. F. (2001). Brand appearances in contemporary cinema films and contribution to global marketing of cigarettes. The Lancet, 357, 29–32. Stacy, A. W., Sussman, S., Dent, C. W., Burton, D., & Flay, B. R. (1992). Moderators of peer social influence in adolescent smoking. Personality and Social Psychology Bulletin, 18, 163–172.

129

Stockwell, T. F., & Glantz, S. A. (1997). Tobacco use is increasing in popular film. Tobacco Control, 6, 282–284. Sussman, S., Dent, C. W., Mestel-Rauch, J., Johnson, C. A., Hansen, W. B., & Flay, B. R. (1988). Adolescent nonsmokers, triers, and regular smokers’ estimates of cigarette smoking prevalence: When do overestimations occur and by whom? Journal of Applied Social Psychology, 18, 537–551. Tickle, J. J., Sargent, J. D., Dalton, M. A., Beach, M. L., & Heatherton, T. F. (2001). Favorite movie stars, their tobacco use in contemporary movies, and its association with adolescent smoking. Tobacco Control, 10, 16–22. U.S. Department of Health and Human Services (1994). Preventing tobacco use among young people: A report of the Surgeon General. Atlanta, GA: Department of Health and Human Services, Public Health Center, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health. Wegener, D. T., & Fabrigar, L. R. (2000). Analysis and design for nonexperimental data: Addressing causal and noncausal hypotheses. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in social and personality psychology (pp. 412–450). New York: Cambridge University Press. Wills, T. A., & Cleary, S. D. (1996). How are social support effects mediated? A test with parental support and adolescent substance use. Journal of Personality and Social Psychology, 71, 937–952.