UC Berkeley

Research has documented that consumers often have imperfect information about the health insurance plans from which they are asked to choose, but we know less about the sources of that information. Given the difficulty in obtaining reliable information from independent sources, consumers may draw on their peers for recommendations. This dissertation investigates the role of social learning in health insurance selection, using longitudinal data from the University of California on plan choices of employees and peers in their department. The data from 2011 to 2016 span a major change in the insurance choice set, which aids in the statistical identification of social learning effects among both incumbent employees as well as new hires. I start by documenting the high similarity in plan choices within peer groups, suggesting the possibility of strong peer effects, and then use a variety of approaches to test for potential confounding from unobserved heterogeneity. I employ a discrete choice conditional logit estimator to formally model plan choice behavior, finding that a 10 percentage point increase in the share of peers who select a particular insurance plan will lead to a 14 percentage point increase in the probability that an individual will choose the same plan. This large effect on plan choice is equivalent to lowering the monthly premium by 18 percent. I then use this model to simulate employer strategies that could exploit social learning to better promote the employer's insurance objectives. For illustration, I conduct counterfactual analyses of incentives to promote adoption of a new consumer-driven insurance plan. At the actuarially fair premium in this setting, demand for a consumer-driven plan is low, and social learning further discourages take-up. However, with sufficient premium subsidies, the model projects that the social learning effects will become positive and can be harnessed by employers to more effectively achieve their cost and insurance coverage goals.