Correspondence
Smoking in the Young
N Engl J Med 1998; 338:472-473February 12, 1998
- Article
To the Editor:
In the October 9 issue, Rigotti et al.1 report on an evaluation of the effectiveness of enforcing laws that ban tobacco sales to minors as a strategy to reduce tobacco use by adolescents. The authors conclude, “We found no meaningful difference in smoking behavior between communities that implemented enforcement programs and those that did not.” This conclusion is intriguing, given the results: the prevalence of 30-day current tobacco use increased significantly over time in the intervention communities (P = 0.01) but not in the control communities (P = 0.93). Contrary to the authors' hypothesis, current tobacco use increased by 2.8 percentage points in the intervention communities as compared with only 0.2 percentage point in the control communities (P = 0.05). Thus, the intervention appears to have had a negative effect. The authors considered this effect to be of “borderline significance” and dismissed it. A negative effect cannot be disregarded on substantive grounds, since other researchers might have predicted such an effect on the basis of a “forbidden fruit” rationale.2 Had the authors found an effect of similar magnitude but in the hypothesized direction, we suspect they would have featured it.
We believe the negative intervention effect should be dismissed because the study used a weak, quasi-experimental design that is subject to numerous threats to internal validity.3 Three intervention communities, in which health departments adopted tobacco-sales regulations, were contrasted with three nearby control communities of similar size and household income.1 Reasonable estimates of the intervention effect can be obtained only if one can rule out plausible alternative hypotheses associated with the assignment to the intervention.3
The intervention and control communities did not have equivalent rates of tobacco use before the intervention. As compared with control subjects, intervention subjects were much less likely to use tobacco at base line (30-day prevalence, 22.5 percent vs. 29.8 percent).1 Furthermore, both groups had lower prevalence rates than those in a 1993 statewide survey of Massachusetts youth (30.2 percent for cigarette use and 9.4 percent for use of smokeless tobacco).4
In the present study, what appeared to be an intervention effect may have been an artifact of statistical regression — extreme observations tend to regress toward the mean when remeasured.3 Fitting alternative statistical models to the data in an attempt to bracket the estimated effect could provide a more valid test of the intervention.3,5 Future research would benefit from stronger study designs, including the use of longitudinal response data and communities that are matched according to base-line tobacco use.
5 ReferencesJoel M. Moskowitz, Ph.D.
Janet Malvin, Ph.D.
University of California, Berkeley, Berkeley, CA 947201
Rigotti NA ,DiFranza JR ,Chang Y ,Tisdale T ,Kemp B ,Singer DE . The effect of enforcing tobacco-sales laws on adolescents' access to tobacco and smoking behavior. N Engl J Med 1997;337:1044-1051
Full Text | Web of Science | Medline2
Glantz SA . Preventing tobacco use -- the youth access trap. Am J Public Health 1996;86:156-158
CrossRef | Web of Science | Medline3
Cook TD, Campbell DT. Quasi-experimentation: design and analysis issues for field settings. Chicago: Rand McNally, 1979.
4
Kann L ,Warren CW ,Harris WA , et al. Youth risk behavior surveillance -- United States, 1993. Mor Mortal Wkly Rep CDC Surveill Summ 1995;44:1-56
Medline5
Reichardt C, Gollub H. Taking uncertainty into account when estimating effects. In: Mark R, Shotland R, eds. Multiple methods in program evaluation. No. 35 of New directions for program evaluation. San Francisco: Jossey–Bass, 1987:7-22.
Author/Editor Response
The authors reply:
To the Editor: We agree with Drs. Moskowitz and Malvin that it would have been preferable to use a stronger study design — ideally, a randomized controlled trial — to address our research question. However, we were evaluating the effect of public policy in the real world, and we used the strongest design available. We could not have recruited Massachusetts communities for a study in which the investigators, not public health officials, decided whether to enforce laws governing tobacco sales. Our compromise, a quasi-experimental design with longitudinal data collection, permitted us to address, in a controlled study, an important and previously unanswered question about the effects of enforcing a tobacco-sales law. The use of this controversial tobacco-control strategy is becoming widespread in the absence of firm data about its effects.
Closer matching of intervention and control communities according to the base-line prevalence of smoking among youth might have been possible, but precise community-level data do not exist. Instead, we matched the communities according to median household income, the best available proxy, since income is correlated with tobacco use. More similar base-line smoking rates would have reduced the concern about regression to the mean. We did not select study communities on the basis of high or low smoking rates, making regression to the mean a less likely alternative explanation of our findings.
What does explain our observations? Enforcement more than doubled compliance rates among merchants, but this change did not have a strong effect, in either direction, on smoking among youth. Tobacco use by adolescents clearly did not fall. Drs. Moskowitz and Malvin raise the possibility that enforcement may actually induce tobacco use among youth, but there was at best a weak effect in this direction. We favor the simpler explanation that the marked improvement in compliance in the intervention communities was not good enough. Stores continued to sell enough tobacco to minors to provide a sufficient supply for them.
Illegal tobacco sales to minors is a stubborn problem. Our results serve as a warning to communities pursuing law enforcement that stronger efforts, such as increasing the frequency of compliance tests or raising penalties for violations, are likely to be needed to stop sales. As our intervention communities learned, there are many political obstacles to such an aggressive approach. Nonetheless, we remain optimistic that vigorous enforcement of the law is possible and can stop the illegal sale of tobacco to children.
Nancy A. Rigotti, M.D.
Daniel E. Singer, M.D.
Massachusetts General Hospital, Boston, MA 02114Joseph R. DiFranza, M.D.
University of Massachusetts Medical Center, Worcester, MA 01655
