UC San Diego

Essays in Workplace Influences on Health Behaviors examines the impact of workplace peers and policies on health behaviors.Chapter 1 examines the impact of coworker fitness on individual fitness and weight status. Using the conditional random assignment of soldiers in the US Army to companies, I find that being assigned to a high fitness peer group increases individual fitness and decreases the likelihood of being obese. The decrease in obesity persists for at least three years following the dissolution of the original peer group. Using an instrumental  variables approach, I isolate the impact of peers’ time varying behaviors and highlight the potential for fitness promoting policies to create positive spillovers. Finally, I show that the peer effect is driven by individuals competing with top performing peers rather than individuals trying to stay ahead of less fit peers.Chapter 2 explores how increased workplace drug testing can lead individuals to substitute away from illicit drugs towards prescription opioids. Using variation in the frequency of unit drug tests in the US Army, we identify the propensity for Company Commander’s to conduct drug testing. We find that soldiers assigned to Company Commanders who conduct more frequent drug testing are less likely to use marijuana and illicit opioids. However, approximately 78% of the decrease in illicit opioid use is offset by an increase in prescription opioid use. In line with a standard deterrence crime model (Becker 1974) we find suggestive evidence that individuals who have a greater expected cost of being discharged from the Army are more likely to decrease illicit drug use when assigned to a high testing commander. Chapter 3 introduces an unbiased alternative estimator to the Leave One Out mean. Our estimator improves upon the Leave One Out mean by eliminating the finite sample bias discussed in Angrist (2014). Simulations show that our estimator is a particular improvement over the Leave One Out mean in settings with small group sizes and a small number of groups. We revisit Chetty et al. (2011) to illustrate the usefulness of our estimator.