Monday, August 22, 2005

100,000 Deaths Per Year Due to Smoking in Movies? The Dangers of Casual Epidemiology

In a column Friday in Reason Online, Jacob Sullum questions the plausibility of the claim that smoking in movies causes approximately 200,000 kids to start smoking each year and will be responsible for 100,000 deaths among these kids per year, and that 50,000 of these deaths could be prevented simply by giving any movie that portrays smoking an R rating. That claim was made by Dr. Stanton Glantz in testimony before Congress in 2004 and can be found on the Smoke Free Movies web site.

There are two major bases for Sullum's questioning of this widely used statistic among anti-smoking advocates.

First, he suggests that there are other factors related to exposure to smoking in movies that could also be related to smoking risk: "The problem with attributing this association to the modeling effect of cinematic smoking is that it's impossible to control for all the differences in personality and environment that make kids more likely to see movies with a lot of smoking in them, which already tend to be R-rated movies."

Second, he suggests that the magnitude of the alleged effect is simply not plausible. If cinematic smoking does cause 52% of all smoking and advertising causes another 34%, than movies and advertising cause 86% of all smoking, leaving only 14% to be explained by other causes: "Methodological difficulties aside, the size of this alleged effect is implausibly large, to put it mildly. Glantz says cinematic smoking accounts for even more real-life smoking than advertising does: 52 percent vs. 34 percent. Is it even conceivable that exposure to movies and advertising causes 86 percent of smoking? That all other factors in life together contribute only 14 percent?"

The Rest of the Story

If I'm reading the statistic correctly, it means that 1 in every 2 kids who start smoking in the U.S. do so because of exposure to smoking in movies and 1 in every 3 kids who start smoking do so because of exposure to traditional advertising. This means that 5 of every 6 kids who smoke do so because of advertising and exposure to smoking in movies, leaving all other causes of smoking to contribute to only 1 of 6 cases of smoking initiation.

And it implies that if no kids ever went out to the movies, the number of smokers would be reduced in half. And if all cigarette advertising were eliminated as well, the overall smoking initiation rate would be cut by 5/6.

It also implies that if all other causes of smoking were eliminated, including parental smoking, parental approval of smoking, smoking in the household, exposure to smoking on television, and exposure to smoking in restaurants, then the number of new smokers would only be cut by 1/6.

It doesn't take a rocket epidemiologist to realize that this is simply implausible. And as a researcher who has for years studied numerous causes of smoking initiation, I don't for a minute believe that all causes of smoking other than advertising and movies explain only one-sixth of all smoking initiation.

Sullum is, I think, insightful in calling out the two major reasons why the statistic is flawed. First, and most importantly, it is simply implausible. Beyond any scientific or methodologic concerns, the implausibility of the statistic is the most important reason why I reject it.

Second, there is a serious methodologic concern, which is the likelihood that smoking in movies is related to a constellation of factors that affect smoking risk. It is extremely unlikely that smoking in movies is single-handedly responsible for 1/2 of all kids who start smoking. Instead, it is more likely the case that exposure to smoking in movies is indicative of a constellation or pattern of social, environmental, and behavioral factors that are directly related to smoking uptake.

Although the study from which the statistic was derived controlled for a number of other factors related to smoking risk -- "grade in school, sex, school, friend smoking, sibling smoking, parent smoking, receptivity to tobacco promotions, school performance, sensation-seeking propensity, rebelliousness, self esteem, parent's education, authoritative parenting, and perception of parental disapproval of smoking" -- there are still likely to be factors that are directly related to exposure to smoking in movies that were not measured (and perhaps, cannot be measured).

In other words, I'm not faulting the study itself - it was an exquisitely well conducted study that did an outstanding job of controlling for potential confounding variables. What I'm suggesting is inappropriate is the casual epidemiology that led to the far-reaching and definitive conclusions from the study that I do not think were warranted.

It is one thing to take a study like this and conclude that exposure to smoking in movies is a significant and substantial cause of smoking and that policy changes, such as requiring movies that portray smoking to carry an R rating, would therefore be effective in reducing youth smoking rates. But it is quite another to take a single study like this and attempt to quantify a precise estimate of the number of kids who start smoking simply due to this single exposure, and then to extrapolate out to conclude that this single exposure will cause a precise number of deaths which could be prevented simply by eliminating this single exposure.

The problem with such casual epidemiology is that it assumes a very simplistic (and probably incorrect) view of the world, in which there is no inter-relationship between various social factors and environmental exposures that kids experience. In reality, life is much richer and more complex, different exposures are intertwined, and the causes of a behavior as complex as smoking are multifactorial, interactive, and collinear.

In casual epidemiology, it may seem that there is a simple line diagram that explains a behavior like smoking. Draw a series of boxes for each contributing factor and then draw a straight line over to a box containing smoking and you have a "casual" epidemiologic model for smoking behavior.

But I think it more likely that there is a "web" of causation - a series of lines that are intertwined, boxes that overlap, lines which lead in both directions. This is not to say that the basic epidemiologic conclusion is incorrect (that exposure to smoking in movies is a cause of smoking); it is simply to say that one needs to exercise caution in taking a single exposure and deriving a precise estimate for the number of cases it causes of such a complex behavior as smoking.

The other major problem with casual epidemiology is that it tends to assume a very simple model of causation - one in which there are independent causes, each of which is necessary and sufficient to cause the relevant outcome. But might there not instead be a more sophisticated pattern of causation, one in which it is the combination of a number of factors that lead to a complex outcome such as smoking. Some of these contributing factors may be necessary to cause the outcome and some may be sufficient, but most are probably neither necessary nor sufficient.

Before I close, I must emphasize that I am not here suggesting that: (1) exposure to smoking in movies is not a substantial cause of smoking; (2) exposure to smoking in movies should not be reduced to lower smoking initiation rates; or (3) an intervention to require movies that portray smoking to carry an R rating is inappropriate.

In fact, I can think of a number of reasons why it may be quite appropriate for movies that portray smoking to carry an R rating, most of which have to do with my concerns as a parent who wants to be able to have some control over what my child is exposed to when the child is away from the home.

But I don't believe that exposure to smoking in movies is single-handedly responsible for half of all kids who start smoking and that requiring an R rating of movies that portray smoking will save 120,000 lives a year.

I'm afraid that the situation is a bit more complex, and we'll have to do a bit more than simply tackle the problem of smoking in movies if we really want to cut the rate of adolescent smoking in half.

UPDATE (August 23, 2005; 8:35 pm): An astute reader and commenter has pointed out that attributable and preventable fractions can add up to more than 100%, assuming a multifactorial theory of causation for smoking initiation (which is of course probable). Thus, my argument above that the claim that exposure to smoking in movies causes 50% of smoking initiation is implausible because it implies that other factors can contribute to only 50% is wrong. My reasoning was based on a single cause assumption. This argument does not, therefore, render the claim that exposure to smoking in movies causes 50% of smoking initiation to be invalid or implausible. Neither does the claim that eliminating all other causes of smoking (beyond movies and advertising) would have to eliminate only 1/6 of all smoking.

For readers who are not epidemiologists, what this all means is that if a behavior like smoking is caused by multiple contributing factors, then if you add up the percentage of smoking caused by each individual factor, it can and will exceed 100%. Take the simple assumption that smoking in movies and parental smoking are both necessary and together are sufficient to cause smoking. In this case, movies cause 100% of smoking initiation and parental smoking causes 100% of smoking initiation. Eliminating smoking in movies will eliminate 100% of smoking, as will eliminating parental smoking. Clearly, these percentages add to more than 100%.

Having said all this, and now apologizing to my readers for having fallen into this single cause thinking, it does not affect my underlying criticism of the claim that exposure to smoking in movies causes half of smoking initiation. The reason is that the claim that is being made is that exposure to smoking in movies is a NECESSARY cause for half of all smoking initiation, not just a CONTRIBUTING cause. And this is what I find to be implausible.

In addition, before concluding that exposure to smoking in movies is single-handedly responsible for 50% of smoking initiation, one would have to rule out the possibility that exposure to smoking in movies is simply a measure of a constellation of other exposures that are related to smoking. It may be, for example, that kids who go out to R-rated movies are also more likely to be exposed to smoking in social settings, and thus, more likely to have a higher perception of community smoking prevalence (a known risk factor for smoking). Since it is difficult to control for all major potential confounding variables, it is critical that one be cautious in making precise claims of the number of lives that would be saved by eliminating a particular factor related to smoking initiation. Drawing such a conclusion from a single study is, in my view, not particularly reasonable.

No comments: