Correlation, causation, and smoking initiation among youths.(Observations)

Expanded consideration of a variety of concepts and methods, from associative learning to econometrics, lends further support to the accumulated consensus that tobacco advertising plays a role, with other factors, in inducing young people to smoke. A point by point rebuttal of issues raised by both Reitter (JAR 43, 1 [2003]: 12-13) and Taylor and Bonner (this issue) makes the case that tobacco advertising is not an exception to the rule: advertising works and it works in part by building primary demand. On a broader, more paradigmatic note, the role of correlation and causation are discussed within a convergence or triangulation framework.

TRIANGULATION AND CAUSALITY

IN CRITIQUING my article "American Media and the Smoking-related Behavior of Asian Adolescents" (Goldberg, 2003), both Taylor and Bonner (2003) and Reitter (2003) have ignored the logic associated with the research strategy of triangulation. Reitter suggests that my purpose in conducting a correlational study was to "supplement and give additional precision" to a study using an experimental design that can lay claim to establishing "causal connections." Rather than addressing the issue at hand, Reitter devotes about half of his comment to a tried and true illustration of spurious correlation. While these examples are always interesting and sometimes amusing, he really misses the point--the presence of a cumulative body of knowledge that allows for triangulation (Webb, Campbell, Schwartz, and Sechrest, 1966).

As discussed in my article, in the absence of a "perfect" study, triangulation allows for studies using a diversity of methods, each with their own strengths and weaknesses to complement one another. This involves a good deal more than one study "supplementing" another. Yes, a correlational study does allow for the possibility of reverse causality or third factors leading to spurious correlation. Long-term prospective or time-series studies attempt to address these issues and, in some measure, they can reduce, if not eliminate, these alternate possibilities. More to the point, complementary experimental studies do allow for the establishment of causality.

Why then are correlational studies needed at all? One only needs to consider how defenders of tobacco advertising respond to experimental studies and their causal findings. They will argue that these are contrived studies conducted in nonnatural settings, with small nonrepresentative samples, using artificial independent and dependent measures. In short, they will highlight not the strengths of the method--the establishment of causality--but its possible weaknesses. By contrast, it is the broader, correlational studies that will often have larger, more representative samples and will study some of the same variables as they occur naturally in the environment. This complementary strength is an important element in a triangulation strategy.

Finally, it should be noted that correlation may not be a sufficient condition for the establishment of causality, but it is a necessary one. The absence of a correlation between the two factors in question would call into doubt the existence of any causal relationship documented between the same actors in an experimental study.

Studies in the domain of tobacco advertising and youth smoking offer a good example of the complementarity referred to above. To juxtapose just one study with the one being discussed, consider the study by Pechmann and Shih (1999), which causally links exposure to smoking in a movie (Reality Bites) to adolescent subjects' intentions to smoke. The study generates a clear causal relationship between exposure to smoking in the movie and youths' smoking intentions. Inherently, as an experiment, the method is limited by the laboratory setting and time frame. Left standing by itself, (these and perhaps other) lingering questions about the external validity of the study might accompany the clear findings of a causal relationship.

A large number of correlational studies, however, including the one presently under consideration, serve to complement and support the causal study. Several content analyses have established that cigarettes have appeared in American movies over the last four decades--on average every 10-15 minutes--and typically are presented in a positive light (Hazan, Lipton, and Glantz, 1994; Ng and Dakake, 2002; Stockwell and Glantz, 1997; Terre, Drabman, and Speer, 1991). The "American Media and Asian Youth" study discussed here considers the pattern between the frequency of attending American movies, not in a lab setting at a single point in time, but in real movie theaters over an extended time period. For a fairly large sampling of Hong Kong youths, a positive relationship is revealed; the more American movies attended, the greater the likelihood of observing smoking-related behaviors. Does this study result in a causal finding? No. Can alternative explanations be offered, yes, especially if the study stood by itself. But we already have a study that indicates there is a causal relationship. Is that study in some ways artificially narrow because of its laboratory setting? Yes. But the present study helps to broaden it by considering the relation ship in a real-world context. These are but two of many studies that buttress one another or triangulate in this fashion and allow the establishment of validity.

Beyond its role in smoking initiation, another function of tobacco advertising is to reinforce behavior already begun. One of the many results of heavy exposure to tobacco advertising is that youths come to view smoking as a prominent and valid part of our society. This "friendly familiarity" (a term coined by Leo Burnett) influences youth before, during, and after smoking initiation.

METHOD

Demand

Suggesting, as do Taylor and Bonner (2003), that somehow the responses to the questionnaire are a function of students responding to the teachers' expectations is something of a stretch. As indicated in the article, teachers were carefully coached on advising students (in the standard manner) that there were no "right or wrong answers." They were further coached to maintain as much distance between themselves and the students as possible. Each student was told that there were no names on the questionnaire, that the responses/ questionnaires were anonymous, and that they were to place their questionnaire into the blank brown envelope provided to each student before returning them. Further, to make the "demand-based" argument suggested by Taylor and Bonner, one would have to assume that only the students who displayed more smoking-related behaviors (but not those who displayed fewer such behaviors) believed that teachers wanted/expected them to link movies and Marlboro advertising (as opposed to advertising for other brands) to smoking. This is a considerable stretch. Similarly, Taylor and Bonner suggest that "retrospective recall" might spuriously account for the results noted in the study. They do not make it clear how retrospective recall, as a spurious …