This dissertation is composed of two chapters. Each chapter presents a study testing a theory from behavioral economics in a health economics setting using field data. The first chapter studies the role of present bias in the choice of health insurance. I analyze the consequences of a policy change that removes deadlines for enrollment in high-quality (5-star) Medicare drug coverage plans (Part D), while maintaining existing deadlines for enrollment in all other plans. Although the goals of the policy were to increase enrollment in 5-star plans and to provide incentives for insurers to improve quality, the removal of deadlines might lead to the opposite. First, rational beneficiaries might wait to enroll in 5-star plans only when a negative health event occurs, which would both decrease enrollment and increase adverse selection. Second, without deadlines, present-biased beneficiaries might procrastinate, which would also lead to a drop in enrollment, driven by an overall increase in inertia. I develop a model to examine these different hypotheses and test its predictions using Medicare administrative micro data for the period of 2009-2012. I employ a difference-in-differences design within a differentiated-product discrete-choice demand framework. My identification strategy takes advantage of the fact that the policy did not actually change enrollment rules everywhere in the United States, as most counties were not within the coverage area of a 5-star provider in 2012, the year the policy was implemented. I have three main findings. First, the policy backfires: it decreases enrollment in the Part D program by 2.55pp from a baseline of 51.76\%, and decreases average market share of 5-star plans by 1.37pp from a baseline of 7.78\%. Second, the policy does not seem to impact adverse selection, suggesting the rational model might not fully account for the results. Third, the removal of deadlines leads to a drop in the probability that a previously enrolled beneficiary switches plans of 3.18pp (baseline 9.08\%), suggesting that at least some Medicare beneficiaries are present-biased. The second chapter studies role of projection bias in mental health treatment decisions. Evidence from psychology suggests that on a bad-weather day, individuals may feel more depressed than usual. If people are not fully able to account for the effect of transient weather, they may take systematically biased treatment decisions. I derive a model of a person considering treatment for depression and show that when projection bias is present, transient weather might influence choice. I use detailed administrative medical records from the MarketScan \textregistered database and daily county-level meteorological data from the National Climatic Data Center. My period of analysis is 01/01/2003 through 12/31/2004. My main analysis focuses on patient behavior during a small interval of time after they have been seen by a physician. I look at how weather influences antidepressant filling decision within patient and only include appointments that involved a major diagnosis of a mental disease or disorder. I find that a one standard deviation increase in the amount of cloud coverage (2.73 oktas) leads to a 0.063 percentage point increase in the probability that a patient fills an antidepressant prescription on appointment day. That is a 1.04\% increase from the 6.07\% baseline. I also find effects associated with snow, rain, and temperature. All effects fade with time and are not significant within seven days of the appointment. Most of the impact of cloud coverage on antidepressant filling is due to an increase on the number of new prescriptions, not an increase in refills. Virtually all the effect happens at the pharmacy, not via mail order. Most regions have similar coefficients associated with cloud coverage, with stronger results in the Northeast and Upper Midwest. Finally, most of the impact happens during Winter.