Smoking and Mortality: New Evidence from a Long Panel

Using data from Framingham Heart Study participants interviewed and examined at frequent intervals over much of their adult lifetime, we discover a relationship between smoking and mortality that differs significantly in magnitude from findings in the medical and epidemiological literature. We attribute the finding of smaller deleterious health impacts of smoking to a failure by previous studies to appropriately account for the non- random nature of lifetime smoking patterns and health histories. We provide causal and consistent estimates of the expected longevity consequences of current smoking, smoking cessation, smoking duration, and smoking experience by jointly modeling individual health heterogeneity, by including supply-side factors such as cigarette advertising and prices, and by allowing for permanent and time-varying unobserved heterogeneity in a flexible yet comprehensive multiple equation joint estimation procedure. We simulate our estimated empirical model under a variety of lifetime smoking patterns, and we compare the resulting mortality differences to the epidemiological literature that treats smoking behavior as random. Our results suggest that difference-in-means statistics significantly overstate the expected longevity loss from various patterns of lifetime smoking. For comparison, the (biased) unconditional difference-in-means in age of death between lifelong smokers and nonsmokers is 9.3 years in our research sample, while simulations from our estimated dynamic model suggest the difference is only 4.3 years. Additionally, our results examine the importance of smoking cessation and relapse avoidance.