Five rural schools in Vermont and New Hampshire participated in a survey that was conducted for a different purpose but contained items relevant to the current hypothesis. A detailed description of the sample and survey methods has been published.7 Briefly, the selection of schools was based on the following criteria: (1) the schools enrolled students in grades 6-12 (ages 10-19 years); (2) the schools were located in rural communities in Vermont or New Hampshire, within a two-hour drive from Lebanon, New Hampshire; (3) schools served communities that were in the lowest quartiles of median family income and percentage of adults who completed high school for each state. Of 13 schools initially contacted and asked to participate, five were surveyed in October 1996: three in New Hampshire (School A, grades 9-12, 469 students; School B, grades 6-7, 169 students; School C, grades 8 – 12, 378 students) and two in Vermont (School D, grades 7-12, 288 students; School E, grades 7-12, 543 students). the populations served by participating schools were not different from non-participating schools.
Participating students were asked to voluntarily complete an anonymous self-administered questionnaire in a classroom or at an assembly. Passive parental consent for administration of this survey at school was obtained by mailing consent forms to students’ homes 1-2 weeks prior to the survey date. Parents were asked to contact the school if they did not want their child to participate in the survey. The Dartmouth University Committee for the Protection of Human Subjects approved the study and consent procedures.
student sample and star preference
Surveys were collected from 1,543 students, representing 79-95% of the students in each school. Eighteen percent of these questionnaires were discarded because the students indicated that they had not been honest, because there were five or more logically inconsistent answers, or because the data on the variables examined in this report were incomplete. complete surveys were obtained from 1265 students. each student’s favorite star was surveyed in the sample by asking the following open-ended question: “who is your favorite film/movie star?” Of 1,236 who responded to this question, 79% were able to identify a favorite actor/actress, 7.9% named a movie or cartoon title (lack of coded response), and 13.2% did not name a favorite actor/actress. favorite actor/actress because the question was open-ended, 228 stars were named in response to the question. assessing smoking status for all the movies by each of these stars would have required us to watch some 2,000 movies. To reduce the number of movies in the sample, we restricted our analysis of stars’ smoking behavior to actors and actresses who were selected by at least five students in the sample and who had appeared in movies within the three years prior to the study ( 1994). to 1996). We exclude another actress, Pamela Anderson, who was probably selected by students who knew her from television and not from movies. these criteria resulted in a sample of 632 students who had selected one of 43 movie stars. the retained students did not differ from those who were excluded with respect to smoking status, grade in school, or exposure to tobacco advertising. they were significantly less likely to have relatives who smoked and were significantly more likely to be male and have below average grades.
validity and reliability
Details of the procedures used to increase the validity and reliability of self-reports have been published.7 We ensured anonymity8 and employed a sham procedure (using a saliva sample)9 to increase the validity of smoking reports. In addition, eighth, tenth, and twelfth graders’ self-reports of tobacco use were similar to those of the nationally representative sample of students surveyed in the 1996 “Future Tracking Survey” (MTFS). retest of all covariates were assessed in a separate survey of 114 students who completed the questionnaire twice, with an interval of five weeks between the two surveys. the κ11 statistic was used to measure the agreement between test responses and test repetition for categorical and ordinal data. no variables needed to be excluded due to unacceptably low κ (< 0.7).12
smoking experience and susceptibility to smoking
We classified students’ smoking status based on two variables that describe attitudes and behavior and have been shown to predict subsequent smoking in prospective adolescent cohort studies. Collins and colleagues13 indicate that the best predictor of future smoking is cigarette consumption at the start of the study. lifetime smoking experience was determined by student responses to two questions: “have you ever tried smoking cigarettes, even a few puffs?” and “how many cigarettes have you smoked in your entire life?” “never smokers” were defined as those who answered “no” and “none”, respectively. “experimental smokers” were those who had smoked less than 100 cigarettes, and “smokers” were those who had smoked 100 or more. Lifetime use of more than 100 cigarettes has been used to classify smokers in studies of adults and is used in the us. uu. to define a person who has been dependent on cigarettes in the past. smoked in the last 30 days.
Pierce and colleagues have shown that a variable describing attitudes toward smoking (termed “smoking susceptibility”) can be combined with measures of experience to more accurately predict post-smoking behavior.15 16 smoking is determined from responses when asked, “how likely is it that the following: (1) will smoke a cigarette in the next six months; (2) I would smoke a cigarette if a friend offered me one.” anyone who cannot definitively rule out smoking in the future by answering “definitely not” for both statements is considered susceptible. prospective studies show that susceptible people who have never smoked are more likely to start using cigarettes in the future. we combined lifetime smoking experience, current smoking, and susceptibility measures into a five-point ordinal smoking status index: never susceptible never smokers, susceptible never smokers, noncurrent experimenters, current experimenters, and smokers. the test-retest reliability κ for this index was 0.95.
variables of smoking in movies
We think that adolescents are more likely to be exposed to contemporary films in which their favorite stars have appeared, and are more likely to pay attention to the behaviors enacted by those stars. thus, we evaluated smoking in all films in the three years prior to the survey (1994 to 1996) in which the favorite stars appeared as main or secondary characters. documentaries, made-for-TV movies, and movies in which the star was the voice of a character (such as in animated movies) were excluded. this resulted in a list of 178 films and 209 roles (some films had more than one actor).
each movie was then viewed to determine if the actors or actresses of interest smoked in the movie. tobacco consumption was measured by a variable as follows: 0 = did not smoke or smoked infrequently (⩽ twice); 1 = smoked frequently (> twice). we used this criterion as a conservative measure of smoking, as only characters shown smoking tobacco multiple times during the film (as opposed to a solitary celebratory cigar, for example) were classified as smokers. The Star Tobacco Use Index is the number of movies during the 1994-1996 sample frame in which a star scored “1” on the tobacco use item. the box office success of each star was measured by the number of movies the star was in that were in the top 50 at the box office in the year the movie was released.
Several characteristics of adolescents and their social environments are known to affect their smoking decisions. In evaluating the relationship between favorite movie star and smoking behavior, we controlled for the effects of other individual and environmental attributes known to be associated with smoking by including these variables as covariates in multivariate analyses. These variables included family and friend smoking, receptivity to tobacco promotions, grade in school, gender, and perceived school performance. Family and friend smoking was measured using Evans and colleagues’ four categories: no friends or family, only family, only friends, family and friends.17 Students were classified as receptive to tobacco promotions if they owned or were willing to use item. School area ethnicity and socioeconomic status (ses) are also known predictors of susceptibility to smoking, but ethnicity and ses in this sample were homogeneous (over 90% of students were white and communities in the sample were in the lowest quartile of ses for the respective states), so these variables were not included as covariates in the model.
We evaluated the association between star tobacco use and adolescent smoking using proportional odds models with smoking rate as the dependent variable. Proportional odds models estimate the probability of being ranked higher in smoking status given preference for stars who portrayed smoking, adjusting for potential confounders. Proportional odds models provide cumulative odds ratios (ORs) that model the probability of being in any higher category on the smoking rate. this method allows for a multilevel dependent variable and preserves information that would be lost by using a dichotomous variable. To assess the robustness of the effect between preferred star smoking behavior and adolescent smoking level, a sensitivity analysis was performed by removing variables from the model or removing individual stars from the analysis. The association between star tobacco use and susceptibility to smoking among never smokers was assessed using multiple logistic regression.