Mortality Risk, Peers, and Risky Behavior

We investigate how a person's decision to engage in risky behavior depends on the behavior's temporal incidence of costs and benefits and on whether or not consumption utility varies with consumption of peers. We define risky goods as goods that deliver immediate utility but that impose costs in two part - a small direct cost when they buy the good; and a (potentially) bigger cost at some future date because when they consume the good, they increase the probability that they will degrade their future health. We incorporate these elements into a model of a person's decision to smoke and consume other goods. In the model, people maximize their (present discounted value of) expected lifetime utility of smoking and consumption subject to presented discounted value of both types of costs.

This structure yields three immediate predictions that we empirically test. We test the predictions that a person will be more likely to start smoking when 1) a greater fraction of his peer group smokes; 2) when he expects to be surrounded by those peers for a longer time; and 3) when he is less likely to survive to pay the costs of consumption that accrue in the future.

To test these predictions, we exploit data on military service and smoking behavior of a sample of men who served in the military between 1933 and 1973 - a period during which the US government required men to serve. We draw data from the Tobacco Use and Veteran's Supplements to the Current Population Surveys (TUS/VA-CPS) and the Panel Study of Income Dynamics (PSID). Both surveys collect retrospectively reported information on military service and lifetime smoking behavior. We use these data to construct lifetime histories of military service and smoking. We validate the data against DoD data on military populations in each year (see Figure 1).

We restrict the sample to men who served in the military and who had not yet started to smoke in the year they first entered military service. We use IV methods to predict whether a man was in active service in a given year. We exploit exogenous changes to the number of months the Department of Defense required men to serve and the substantial variation in the mortality risk.

To give some sense of that variation, we plot, the mortality rate of military men from 1933 to 1980 and the share of men in our sample by the year they first entered military service. In the TUS-CPS sample, we also have and exploit data on whether men served in combat during the years of their military service.

Results support the model's predictions. Men are much more likely to start smoking in years that mortality rates were higher, when a greater fraction of their peers smoke, and when they were required to serve longer.