BACKGROUND. Previous studies showed an association between viewing of smoking in movies and initiation of smoking among adolescents. However, all studies except one were cross-sectional, and none updated movie smoking exposure prospectively or assessed its influence on children.
METHODS. We enrolled elementary school students, 9 to 12 years of age, in a longitudinal study to assess the influence of movie smoking exposure on smoking initiation among children. Movie smoking content was coded for the most popular movie releases; exposure was assessed by asking children which movies they had seen, on the basis of unique lists of 50 movies sampled randomly from top box office hits and video rentals. Data collection occurred in 3 waves (the baseline survey and 2 follow-up surveys), ∼1 year apart. Movie lists were updated for each data collection wave, to reflect recent releases. Movie smoking exposure was analyzed in relation to smoking initiation by the end of the study period.
RESULTS. Approximately 80% of the children's smoking exposure occurred through movies rated G, PG, or PG-13. Children's movie smoking exposure predicted smoking initiation significantly, after adjustment for multiple covariates including child and parent characteristics. The relative risks were 1.09, 1.09, and 1.07 for a 1-decile increase of movie smoking exposure measured at the baseline, second, and third data collection waves, respectively. The adjusted attributable risk of smoking initiation attributable to movie smoking exposure was 0.35.
CONCLUSION. Our study, which is the first to enroll children in elementary school and to update movie smoking exposure longitudinally, indicates that early exposure has as much influence on smoking risk as does exposure nearer the outcome. Overall, movie smoking may be responsible for at least one third of smoking initiation for children in this age group.
Over the past several years, researchers have demonstrated through empirical studies what tobacco companies have known for decades—that celebrity use and movie images of smoking can be even more powerful than commercial advertising.1 A 1989 market research study for Phillip Morris concluded “most of the strong, positive images for cigarettes and smoking are created by cinema and television.”2 Almost 2 decades later, this statement still rings true.3,4 Unfortunately, children and adolescents, many of whom are dedicated movie viewers, are particularly vulnerable to these persuasive portrayals of smoking.5 A number of epidemiologic studies of adolescents found direct evidence that viewing more movie smoking increased the likelihood of smoking initiation.1 Those studies included our previous reports based on cross-sectional6 and follow-up7 evaluations of a regional sample of adolescents and recent cross-sectional findings from a nationwide sample of adolescents.8 A recent meta-analysis of peer-reviewed studies on tobacco marketing and entertainment media confirmed that movie smoking is at least as influential as cigarette advertisements and marketing campaigns in promoting smoking initiation among adolescents.9
Although the peak period of smoking initiation is during adolescence, the desire to smoke may develop at much younger ages. Early exposure to social situations and media that depict smoking may foster positive attitudes toward smoking behavior,10–12 but whether such exposure also predicts smoking behavior remains unknown. Previous studies of movie smoking exposure focused on adolescents, and most participants were between 10 and 14 years of age at enrollment.6–8 In addition, exposure to movie smoking was measured at only one time point; consequently, it was not possible to assess whether early movie smoking exposure was as influential as later exposure with respect to risk of smoking initiation. The current study was designed to evaluate the influence of children's exposure to movie smoking over time and to assess the different effects of early exposure and exposure occurring closer to the time of smoking initiation.
This research was approved by the Committee for the Protection of Human Subjects at Dartmouth College. Child participants were identified through schools in New Hampshire and Vermont. Schools were selected randomly from all schools containing grades 4 through 6 (N = 559), stratified according to state and the number of students enrolled. Twenty-six schools participated in the study, representing 30% of those contacted. Each participating school received a stipend of $500 to $750.
Data collection proceeded in 3 phases, that is, a baseline survey and 2 telephone follow-up surveys (waves 2 and 3 of data collection). Baseline data collection (completed between 2002 and 2003) included both an in-school, self-administered, written survey and a supplementary telephone interview.13 Children who participated in the in-school survey provided their names and telephone numbers on a separate sheet of paper, which allowed us to contact their parents for consent to interview the children by telephone. If ≥2 siblings completed the in-school survey, then we selected randomly 1 child from each household, to ensure unique parent/child dyads. The supplementary baseline telephone interviews with the children were conducted an average of 8.4 weeks (median: 7.6 weeks) after the in-school survey. Wave 2 of data collection took place an average of 47.5 weeks (median: 46.6 weeks) after the baseline data collection, and wave 3 took place an average of 53.5 weeks (median: 53.3 weeks) after wave 2. We also conducted baseline telephone interviews with the children's parents; these took place after the children's interviews, to minimize the possibility of influencing the children's responses. Whenever possible, we interviewed the child's mother. The child and parent telephone surveys were administered by trained interviewers, using an individualized computer-assisted telephone interview system. To protect confidentiality, child participants indicated their answers by pressing numbers on their telephone.
The outcome of interest, namely, lifetime smoking experience, was assessed at baseline and at each of the 2 follow-up interviews by asking children, “How many cigarettes have you smoked in your life?” Respondents could answer none, just a few puffs, 1 to 19 cigarettes, 20 to 100 cigarettes, or >100 cigarettes. The children's exposure to smoking in movies was assessed at baseline by asking each student to indicate which films he or she had seen, from a unique list of 50 movies. By using a previously described method for estimating movie smoking exposure,6–8 each unique baseline list was selected randomly for individual surveys from 550 popular contemporary movies, representing each year's top 100 box office hits for the 5.5 years preceding the child's survey. The original list of 550 movies was updated twice during the baseline data collection period, to ensure that all enrollees received lists representing the most recent 5.5-year period; overall, the unique lists were drawn from top box office hits representing 1997 through the first 6 months of 2003. We stratified the random selection of movies so that each list of 50 had the same distribution of ratings as the larger sample of top box office hits (41.1% were R-rated, 40.0% PG-13, 14.2% PG, and 4.8% G). For waves 2 and 3, the unique lists of 50 movies were selected on a rolling basis from the top 100 box office hits plus the top 100 video rentals for the 12-month period preceding the survey; because of overlap between the box office hits and video releases, the sampling frame for each 12-month period consisted of <200 movies. Overall, the sampling frame for the entire study consisted of 899 unique movie titles. The vast majority (96.4%) of movie titles about which individual children were asked across the 3 survey waves were unique titles, and <1% of movies were reported more than once by a child. On average, for an individual child, <2% of movie smoking exposure measured at wave 2 overlapped with baseline measures, and none of wave 3 exposure overlapped with baseline measures.
The detailed methods used to code movies for smoking occurrences were described previously.3,4 Briefly, trained coders counted the number of smoking occurrences in every movie included in the sampling frames. A smoking occurrence was defined as any use or handling of tobacco by a major, minor, or ancillary (one or more) character in a new scene. For each child, exposure was calculated by summing the number of smoking occurrences in the movies the respondent had seen. Using questions adapted from previously validated instruments, we also measured variables that could potentially confound the association between movie smoking exposure and adolescent smoking initiation. The information obtained through the child's school survey included child's age, gender, race, and school performance. We also assessed self-regulation (4 items), sensation-seeking (4 items), self-esteem (4 items), rebelliousness (6 items), maternal responsiveness (5 items), and maternal demandingness (9 items). The personality characteristic and maternal responsiveness scales14 were adapted from our previous study6 through reduction of the number of items in each scale to accommodate the younger age of participants. Children used a 4-point scale (from 0 to 3) to indicate how well each item described themselves or their mothers. Summary scores were created by summing the children's responses to each item; higher scores indicated more of the characteristic. Maternal responsiveness and child personality characteristics were assessed through the child's baseline school survey. Maternal monitoring and friend smoking status were assessed through the child's baseline telephone interview, and friend smoking status was updated at each wave. Variables obtained through the parent baseline survey included parent race, education, and smoking status.
The initial sample consisted of 2627 children who completed the in-school and supplemental telephone baseline surveys. Of those, 2499 children had not tried smoking previously and were eligible for follow-up evaluations. Wave 2 data collection was accomplished for 2354 (94.2%) of the 2499 eligible children, and complete follow-up data were obtained for 2255 (90.2%). Baseline telephone surveys were completed for 98.4% of the children's parents. The characteristics of the follow-up study participants were comparable to those described in our previous report for 2606 parent/child dyads, including the baseline smokers.13 Briefly, in this study, 94.7% of participating parents were mothers, 4.3% were fathers, and 1% were other primary caregivers. Most parents (90%) reported their race as white, 30.4% had completed high school (or less), 38.4% had some college education, and 31.2% had a bachelor's degree or more, similar to the underlying population of adults in New Hampshire and Vermont. Approximately one half (50.1%) of the children were male; 75.9% reported their race as white, 7.2% as Latino, and 3.9% as black, Asian, or Native American. The remaining children reported themselves as mixed race, which may reflect ancestry of mixed national origin (eg, Italian and Irish), rather than mixed race.
To assess the extent of possible bias introduced by incomplete follow-up data, we used logistic regression to examine the relationship between dropping out of the study and baseline characteristics of parents and children, treated as dichotomous outcomes (Table 1); t tests were used to assess differences in mean values for the participant and dropout groups. Children who dropped out of the study were more likely to report a race other than white and to have school performance below excellent. They were also more likely to have a parent who smoked or whose highest educational level was a high school diploma or less. The mean movie smoking exposure was slightly higher for dropouts, compared with those who remained in the study. Dropping out of the study was associated with rebelliousness but was not associated significantly with other child characteristics, including gender, age, sensation-seeking, self-regulation, and self-esteem. Dropping out also was unrelated to the child's assessment of maternal monitoring and responsiveness.
The main goal of our analysis was to assess movie smoking exposure over time in relation to smoking initiation by wave 3. Movie smoking exposure, estimated as described above, was expressed as a percentile and treated as a continuous variable in the analyses. The relative risks (RRs) for smoking initiation are shown per 1-decile increase in movie smoking exposure for ease of presentation. The outcome of initiating smoking was defined as any smoking (“just a few puffs” or more) and treated as a binary variable (yes or no). Children who reported any cigarette smoking in a follow-up survey were classified as having initiated smoking. We used Poisson regression to compute RRs and 95% confidence intervals (CIs) for the relationship between movie smoking exposure and initiation of smoking by wave 3. The analyses considered the effect of cumulative movie smoking exposure and the influence of exposure measured at individual waves of data collection. An overdispersion parameter15 was used to adjust for possible clustering according to school, because failure to adjust for clustering might lead to falsely low P values. Exposure to movie smoking and having a friend who smoked were updated at each data collection wave. All other measures were taken from the baseline survey. The analysis did not consider exposure that occurred after a participant reported smoking initiation. In the multivariate models, RRs were adjusted for the following covariates, using the cutoff points shown in Table 2: age, gender, race, school performance, sensation-seeking, rebelliousness, self-regulation, self-esteem, having a friend who smokes, parent education, parent smoking status, maternal monitoring, and maternal responsiveness.
Preliminary analyses indicated substantial correlation of movie smoking exposure across the waves of data collection, precluding simultaneous assessment of exposure waves in a multivariate model. To isolate the effect of each wave's movie smoking exposure, we examined the effect of residual movie smoking exposure. The residual effect of smoking exposure was estimated as the difference between the actual movie smoking exposure at a follow-up data collection phase and the movie smoking exposure predicted for that wave. For a given wave, predicted movie smoking exposure was a function of the association between covariates and movie smoking exposure in the previous wave. This approach allowed direct comparison of movie smoking exposure coefficients for each wave in a single model while averting the problem of collinearity between measures of movie smoking exposure at each wave.
We also estimated the proportion of smoking initiation attributable to excess movie smoking exposure. In these calculations, we defined excess movie smoking exposure conservatively, as exposure above the 25th percentile. The attributable risk (AR) estimates were calculated by assessing the reduced smoking risk associated with decreasing a child's movie smoking exposure from its actual level to the 25th percentile. For example, if a child's exposure was at the 75th percentile, then the analysis considered the risk reduction accomplished by decreasing the child's exposure to the 25th percentile. Using this approach, the covariate-adjusted AR was estimated for each child, and the overall AR was calculated by summing these estimates over the cohort. The analysis was repeated to assess the reduced smoking risk associated with decreasing a child's movie smoking exposure from its actual level to the 10th percentile. Because of the nonlinear relationship of AR to covariates, we used bootstrap simulation procedures to estimate SEs.
The children in the study saw 36.8 ± 0.36 (mean ± SE) of the 150 movies on their 3 unique lists, including 13.8 ± 0.09 G/PG-rated movies, 17.3 ± 0.21 PG-13 movies, and 5.3 ± 0.13 R movies. Through these movies, the children were exposed to an average of 149.6 (SE: 2.36) smoking occurrences, including 40.7 occurrences at baseline, 53.4 occurrences at wave 2, and 55.9 occurrences at wave 3. Of the total occurrences seen, 18.9% were from G/PG movies, 60.4% were from PG-13 movies, and 20.7% were from R movies. On the basis of the 3 lists of movie titles, children in the lowest decile saw an average of 31 smoking occurrences, those in the 25th percentile saw 67 occurrences, those in the 75th percentile saw 209 occurrences, and those in the 90th percentile saw 300 occurrences.
By the third data collection wave, 217 children (9.6%) had initiated smoking; of those, 86 had initiated smoking by wave 2. Of the 217 children who reported smoking by wave 3, 159 children had smoked just a few puffs, 39 had smoked 1 to 19 cigarettes, and 19 children had smoked >19 cigarettes. None of the children reported smoking >100 cigarettes.
In Table 2, child and parent characteristics measured at baseline are shown in relation to smoking initiation by the third wave of data collection. The analyses identified several potential confounders of the association between movie smoking exposure and smoking initiation. Baseline child characteristics including female gender, older age, lower school performance, having a friend who smokes, high sensation-seeking, high rebelliousness, low self-regulation, and low self-esteem were associated significantly with smoking. Children were at higher risk if their parents did not complete college and if ≥1 parent smoked. Low maternal responsiveness and low maternal monitoring also were associated with initiation of smoking. Terms for these covariates were included in the multivariate models.
Figure 1 shows the unadjusted relationship between the percentile of movie smoking exposure at each wave of data collection and the risk of initiating smoking by wave 3. Substantial overlap of the curves suggested that baseline movie smoking exposure predicted initiation of smoking as well as movie smoking exposure measured at subsequent waves. Because age is potentially the strongest confounder in the relationship between exposure and smoking initiation, we assessed whether the influence of baseline movie smoking exposure might be related to the children's ages at study enrollment (Fig 2). The results suggested that the association between percentile of baseline movie smoking exposure and smoking initiation was most apparent for children who were oldest (12 years of age) at the time of study entry. A consistent relationship between movie smoking exposure and initiation of smoking was apparent for the oldest children, whereas risk began to accelerate above the 60th percentile for the youngest children (9 years of age). The possible interaction between baseline age and exposure, however, was not statistically significant.
Movie smoking exposure was correlated across waves of data collection (r = 0.47 for waves 1 and 2, r = 0.41 for waves 1 and 3, and r = 0.61 for waves 2 and 3); therefore, it was impractical to use a multivariate model to estimate simultaneously the effect of each exposure wave. Consequently, we used 3 separate models, each containing terms for the covariates identified in Table 2, to assess the influence of each wave of exposure on initiation of smoking by the end of the follow-up period. The results showed similar effects for each wave of movie smoking exposure (Table 3). The covariate-adjusted RRs for a 1-decile increase in exposure were 1.09 (95% CI: 1.03–1.15), 1.09 (95% CI: 1.03–1.16), and 1.07 (95% CI: 1.00–1.14) for exposure measured at baseline, wave 2, and wave 3, respectively. These analyses suggested that the influence of baseline exposure was similar to that of later exposure in predicting smoking initiation. We also assessed the influence of cumulative movie smoking exposure measured across waves of data collection. The RRs were 1.11 (95% CI: 1.04–1.17) for a 1-decile increase of exposure accumulated by wave 2 and 1.09 (95% CI: 1.02–1.16) for exposure accumulated by wave 3 (Table 3). These findings also indicated similar risks associated with baseline exposure and exposure occurring later in the follow-up period.
An additional analysis was undertaken to isolate the effects of all 3 measures of movie smoking exposure in a single multivariate model, while averting the problem of collinearity. In this approach, we estimated residual movie exposure, defined as the difference between actual movie smoking exposure at a follow-up data collection phase and the exposure predicted for that wave on the basis of the covariate-exposure relationship in the previous wave. This analysis indicated that baseline exposure was a significant predictor of initiation of smoking by wave 3 (P = .02) and that prediction was not improved by the additional exposure contributed at wave 2 (P = .24) or wave 3 (P = .47).
To estimate the contribution of movie smoking exposure to smoking initiation, we examined the influence of excess movie smoking exposure, defined as exposure above the 25th percentile, on smoking initiation by wave 3. The covariate-adjusted results are shown in Fig 3. Similar effects were seen for the 3 waves of data collection; that is, the predictive value of excess movie smoking exposure measured at wave 2 or wave 3 was similar to that assessed at baseline. The AR was computed as an estimate of the proportion of children who initiated smoking as a result of seeing smoking in movies. Using exposure below the 25th percentile as the reference group, the AR was 0.35 (95% CI: 0.16–0.53). When we repeated the analysis by using exposure below the 10th percentile as the reference group, the AR was 0.46 (95% CI: 0.11–0.70).
This is the first study to focus on elementary school children, to update movie smoking exposure prospectively, and to assess the relative importance of movie smoking exposures occurring early in childhood versus those occurring nearer to the time of smoking initiation. Consistent with previous studies of adolescents,6–8 our data showed that movie smoking exposure was associated with an increased risk of smoking initiation. Importantly, our study also showed that the earliest measure of movie smoking exposure was similar in importance to later measures. This finding was upheld consistently in separate analyses that isolated the effects of each exposure wave, examined cumulative exposure, or assessed the effects of residual exposure. Therefore, our data indicate that movie smoking exposure occurring during early childhood is as influential as exposure that occurs nearer to the time of smoking initiation.
Because our baseline movie list included movies that were released up to 5.5 years before the survey, children who were 12 years of age at baseline were potentially as young as 6 to 7 years of age when first exposed to smoking through movies on their list, and the youngest children, enrolled at 9 years of age, were potentially of preschool age at the time of exposure. Therefore, our findings are consistent with a durable influence of early movie smoking exposure on young children. A latent effect of early exposure on later smoking behavior may also be operative. A previous study based on a diverse sample of schoolchildren, 11 to 16 years of age, in London, England, showed that children who tried even a single cigarette were at greater risk to start smoking at a later age, despite an interval of ≥3 years of nonsmoking.16 Because the baseline measure covered an extended time period, it is also possible our findings reflect the longer time frame of exposure. Exposures measured at each wave were correlated strongly; therefore, we were limited in our ability to detect the extent to which changes in movie smoking exposure at each wave influenced the risk of smoking initiation. Also, smoking portrayals in movies might have changed during the period of time covered by our study, a possibility that would not be captured by our exposure measure. However, because the RRs from the separate multivariate models were similar for each wave, this seems an unlikely explanation of our findings. As a cautionary note, although this study supports the importance of early exposure, our results should not be interpreted as suggesting that movie smoking exposure occurring at a later age is not detrimental. The importance of early exposure observed here might reflect a delayed influence coupled with a short follow-up period. Later exposure also might have a delayed influence that could be appreciated with extended follow-up evaluation.
Our estimate of AR indicated that 35% of smoking initiation in this age group is attributable to movie smoking exposure. This estimate is lower than that observed in our previous longitudinal study of adolescents (AR: 0.52; 95% CI: 30.0–67.3)7 but is within the 95% CI. The lower AR observed here may reflect the lesser exposure observed in the younger children. Adolescents in our previous longitudinal study saw an average of 16 of the 50 movies (99 smoking occurrences),7 whereas children in the present study saw an average of 10 of the 50 movies (41 smoking occurrences) on their baseline lists. In comparisons of the highest and lowest quartiles of movie smoking exposure, the fully adjusted RR was 2.7 in the study of adolescents7 and 2.3 in the present study. Our AR is similar to that from a cross-sectional analysis of baseline data collected from a national sample of adolescents (AR: 0.38; 95% CI: 0.20–0.56).8 In that study, children were 10 to 14 years of age at baseline, similar to the ages of our study participants at the end of the follow-up period.
Although the peak period for smoking initiation is during adolescence, our data suggest that preventive efforts need to begin much earlier. We and others have suggested several approaches that parents can use to decrease their children's risk of smoking, including restricting their viewing of R-rated movies, which have the most smoking occurrences.13,17–20 The present study findings suggest that parents also need to consider the smoking content of movies with lower ratings. Most (59%) of the movies seen by these children in elementary school were youth-rated (G, PG, or PG-13), and youth-rated movies accounted for most (79%) of the movie smoking exposures. Similarly, a recent report showed that more than one half of the tobacco occurrences in the top 100 movies of 2004 appeared in youth-rated movies.4 Because smoking is not a specific criterion in the Motion Picture Association of America movie rating system, it would be very difficult for parents to try to identify and to select youth-rated movies without smoking. A more-practical solution to reduce children's exposure to movie smoking would be to eliminate smoking from youth-rated movies.
Funding for this study was provided by the National Cancer Institute (grant R01 CA94273). The funding agency was not involved in study design or implementation, statistical analyses, or interpretation of findings.
We thank our research team for their efforts on this study, in particular, Susan K. Martin for data collection and Aurora Matzkin and Jennifer Gibson for data analysis. We are especially grateful to the children, parents, and schools whose participation made this study possible.
- Accepted July 12, 2007.
- Address correspondence to Linda Titus-Ernstoff, PhD, MA, HB 7927, Dartmouth-Hitchcock Medical Center, One Medical Center Dr, Lebanon, NH 03756. E-mail:
The authors have indicated they have no financial relationships relevant to this article to disclose.
- ↵Charlesworth A, Glantz SA. Smoking in the movies increases adolescent smoking: a review. Pediatrics.2005;116 :1516– 1528
- ↵Morris P, ed. Cigarette Marketing: A New Perspective. London, England: Kelly Weedon Shute Advertising; 1989
- ↵Worth K, Tanski S, Sargent J. Trends in Top Box Office Movie Tobacco Use, 1996–2004: First Look Report 16. Washington, DC: American Legacy Foundation; 2006
- ↵Sargent JD, Beach ML, Dalton MA, et al. Effect of seeing tobacco use in films on trying smoking among adolescents: cross-sectional study. Br Med J.2001;323 :1394– 1397
- ↵Sargent JD, Beach ML, Adachi-Mejia AM, et al. Exposure to movie smoking: its relation to smoking initiation among US adolescents. Pediatrics.2005;116 :1183– 1191
- ↵Dalton MA, Adachi-Mejia AM, Longacre MR, et al. Parental rules and monitoring of children's movie viewing associated with children's risk for smoking and drinking. Pediatrics.2006;118 :1932– 1942
- ↵Jackson C, Henriksen L, Foshee VA. The Authoritative Parenting Index: predicting health risk behaviors among children and adolescents. Health Educ Behav.1998;25 :319– 327
- ↵McCullagh P, Nelder JA. Generalized Linear Models. 2nd ed. New York, NY: Chapman and Hall; 1989
- ↵Fidler JA, Wardle J, Broderson NH, Jarvis MJ, West R. Vulnerability to smoking after trying a single cigarette can lie dormant for three years or more. Tob Control.2006;15 :205– 209
- Pierce JP, Farkas A, Evans N. Tobacco Use in California, 1992: A Focus on Preventing Uptake in Adolescents. Sacramento, CA: California Department of Human Services; 1993
- Sargent JD, Beach ML, Dalton MA, et al. Effect of parental R-rated movie restriction on adolescent smoking initiation: a prospective study. Pediatrics.2004;114 :149– 156
- ↵Sargent JD, Dalton MA. Does parental disapproval of smoking prevent adolescents from becoming established smokers? Pediatrics.2001;108 :1256– 1262
- Copyright © 2008 by the American Academy of Pediatrics