Mitigating Risk: Smartphone Notifications, Adaptive Surveying, and Genetics.

Abstract

Modern information technologies have revolutionized how we can study risk taking. I leverage datasets produced by three of these technologies to investigate the mitigation of risk associated with obesity, saving for retirement, and smoking. First, I investigate how smartphone notifications, namely those sent by wearable activity trackers and associated mobile "apps," can mitigate risks associated with unhealthy living. The notifications suggest temporary, modestly ambitious goals to randomly treated users. They frame not meeting the temporary goal as a loss, tapping into loss aversion, and thereby promoting physical activity in the short-term (on the order of 0.10 more miles of walking for 2/3 step interventions) and reported sleep in the medium-term (on the order of a few more minutes for the sole sleep intervention). Second, I focus on mitigating risks associated with Americans saving inadequately or incorrectly for retirement. Framing biases permeate measurements of risk preferences, clouding what to assume in constructing portfolios and default policy rules, and potentially causing large welfare losses. We develop an adaptive surveying procedure to parse framing biases from measures of risk aversion by allowing subjects to reason through inconsistencies among differently framed (but axiomatically equivalent) hypothetical retirement savings plans. The procedure induces an overall reduction in decision errors and an increase in consistency across frames. It also engenders mild convergence in population estimates of mean risk aversion among frames, suggesting resultant "reasoned" preferences better represent true, underlying preferences. Third, using data from the Health and Retirement Study (HRS), I explore whether genetic predisposition can mitigate the extent to which people response to changes in cigarette excise taxes. I focus on two sorts of genetic predisposition, namely to high daily cigarette consumption and high educational attainment. I construct these predisposition measures, or polygenic scores, by performing out-of-sample prediction on the HRS using effect size estimates associated with millions of genetic markers from past, large genome-wide association studies (which exclude HRS). Both polygenic scores are predictive of smoking status and cigarettes per day. However, along both consumption margins and with respect to either polygenic score, I find no evidence of heterogeneity in responsiveness to cigarette excise taxes.