Monday, November 07, 2005

On Movies as a Cause of 38% of Youth Smoking: The Limits of Causal Inference from a Cross-Sectional Study

A study being released today in the journal Pediatrics concludes that smoking in movies is a major cause of youth smoking, responsible for 38% of all youth smoking initiation in the United States. The conclusion is based on the results of a cross-sectional telephone survey of 6,522 adolescents ages 10-14 years old. Exposure to smoking in movies was assessed by ascertaining whether youth respondents had seen 50 random movies chosen from a list of 532 first-run movies. The association between smoking in movies and youth smoking status was then examined, controlling for a number of potential confounding variables (including parental and peer smoking). There was a clear dose-response relationship between the estimated exposure to smoking in movies and the odds of a respondent being a smoker. Based on this relationship, the paper concludes that seeing smoking in movies is responsible for 38% of smoking initiation in the United States.

The basic conclusion of the study is as follows: "Our study suggests that exposure to movie smoking is a primary independent risk factor, accounting for smoking initiation in more than one-third of US adolescents 10 to 14 years of age, and provides additional scientific support for public health programs aimed at reducing adolescent exposure to movie smoking."

A press release publicizing the research findings concludes: "The research, supported by the National Cancer Institute, suggests that exposure to movie smoking accounts for smoking initiation among over one-third of U.S. adolescents."

The Rest of the Story

While I agree that there is strong evidence that exposure to smoking in movies is an independent risk factor for smoking initiation and that reducing youth exposure to smoking in movies is therefore a good idea, I do not particularly find it scientifically strong to take the results of this cross-sectional study and make a claim that 38% of kids who start smoking do so because they saw people smoking in films.

There are three major potential problems with the validity of such a claim based on this specific research design.

First, the study cannot establish whether the youths who started smoking actually saw the smoking in movies before they took up smoking. It is entirely possible that a large number of these youths actually started smoking prior to the time when they were exposed to the bulk of the smoking depictions from these movies.

In fact, the study itself acknowledges this possibility: "the present study does not preclude the possibility that smoking initiation preceded movie smoking exposure in this sample."

Well if the present study does not preclude the possibility that smoking initiation preceded movie smoking exposure, then I don't see how the results of the study can possibly be used to conclude that exposure to smoking in movies causes 38% of all adolescent smoking initiation.

The problem here, as I see it, is not in making a definitive conclusion that exposure to smoking in movies contributes to youth smoking initiation. That's not an unreasonable conclusion to draw from the evidence.

What I do find unreasonable, however, is quantifying the effects of movie exposure precisely in terms of an estimate of the exact proportion of kids who start smoking from seeing movies, when in fact the data come from a cross-sectional study in which it is not even clear that the bulk of movie smoking exposure even preceded the initiation of smoking.

It is entirely possible that kids who smoke tend to be more likely to view films that tend to contain more depictions of smoking. It's not only possible - it's likely.

The second major problem is that in this type of cross-sectional study, it is very difficult to rule out the confounding effects of factors that are likely to be related both to viewing movies with more smoking in them and smoking initiation. The paper does a tremendous job of trying to account for as many of these factors as possible, but I would be extremely reluctant to quantify the precise number of youths who start smoking due to seeing smoking in films based on this cross-sectional study.

The third major problem is that exposure to smoking in movies is likely to represent a proxy measure for a wider constellation of media-related exposures to smoking that likely all contribute to the smoking initiation process. These include magazine advertising, outdoor advertising (especially on the exterior of stores and point-of-sale advertising within stores), exposure to smoking on television, etc.

I don't think it is valid to isolate the specific impact of exposure to smoking in movies and claim that this exposure alone is the cause of youth smoking in these 38% of youth smoking cases. It is likely that a much wider array of exposures is contributing to the initiation process. Exposure to smoking in movies may certainly be one of them, but it is not the only one, and I do not believe that from these cross-sectional data which didn't measure any of these other exposures, one can tease out the smoking in movies and claim that it is the precise (and sufficient) cause of the observed smoking behavior.

Why is this important? After all, couldn't one argue that since the ultimate effect of this research and any resulting policy action will be to reduce exposure to smoking and decrease smoking (albeit by a lower amount than 38%), it is ultimately not a problem?

I think it is a problem, however, for two reasons.

First, I think that despite all this scientific work being for a great cause, it ultimately hurts the science base for tobacco control in general to have specific claims being made from studies that do not support those claims. The fact that a cross-sectional study that was unable to determine the timing of exposure to smoking in movies and the initiation of smoking and was unable to examine the range of smoking-related media exposures that youths experience was used to make a precise quantitative estimate of the specific reason why youths start to smoke tends to undermine the overall science of making population-based risk estimates in tobacco control.

It is only an observation, but it certainly seems to me that the rigor required before we start making these kinds of precise quantitative claims has decreased over the years in which I have been doing research in this field. Ultimately, I guess that I am concerned somewhat about the credibility of tobacco control research findings among the public if they are continually exposed to these types of quantitative claims being made from research designs that are simply not "designed" to produce such claims.

Second, if, for the reasons I have outlined above, the claim is an overestimate, then we will fail to see the expected decline in youth smoking if we do address the problem definitively. So even if we were to ban all smoking in movies tomorrow, I very much doubt that we would subsequently see a 38% reduction in youth smoking initiation.

The narrow focus on one source of exposure to smoking in the media, stimulated by the mistaken belief that this single source is the primary reason why kids smoke, might actually lead public health practitioners to take their eyes off of other important exposures that influence the smoking initiation process. If it is smoking in movies, and not advertising of smoking in magazines or in stores, smoking on television, and smoking in other media that is causing such a huge proportion of kids to smoke, then it makes perfect sense to ignore these other media and to focus, instead, solely on smoking in movies.

I think that would be a severe mistake, as I believe it would result in very little impact, if any, on youth smoking initiation. I think we need to take a broader view and try to address the constellation of media-related exposures that are most likely working together and all contributing to the smoking initiation process. This is not to say that focusing, in part, on smoking in movies is not entirely appropriate. It is only to demonstrate why I think that there is a danger in making precise quantitative claims of the specific impact of smoking in movies in light of the methodologic problems that I have noted.

The rest of the story suggests that while exposure to smoking in movies is likely a significant factor that contributes to smoking initiation, the scientific evidence does not support a precise conclusion that 38% of youth smoking is caused by smoking in movies (and therefore that if smoking in movies were eliminated, youth smoking would drop by 38%). In fact, I do not believe that smoking in movies is the cause of 38% of kids starting to smoke. I think it is likely a significant factor, but that it likely works with other media-related exposures to change the perception of the prevalence and social acceptability of smoking, which in turn, influence the initiation process.

No comments: