Information Design for Epidemic Containment

Abstract

In this thesis, we study the possibility of nudging interacting individuals on a network, using strategic information disclosure, to make pro-social decisions during epidemics. By posing our problem in the framework of Bayesian persuasion, we formulate a model that captures the decision of choosing the optimal information disclosure policy at the level at the government, and the individuals' decisions of choosing effort levels to stay safe during the epidemic, given the actions of their neighbours and the information communicated to them by the government. We study the optimal information disclosure policy for two objectives of the government: maximizing the expected societal aggregate effort exerted to keep itself safe during the epidemic, and maximizing the probability of a randomly selected individual being safe during the epidemic. We find that the optimal policy depends on the risk associated with the probability of being safe during epidemics and some of its associated functions. These functions have an implicit dependence on the public's belief about the epidemic state. We find sufficient conditions, which do not depend on the public's belief, under which the extreme policies of full information disclosure and no information disclosure are optimal. Our analysis also provides insight into other forms of intervention design, such as mechanism design and network design, which we discuss in this thesis. Having focused on publicly announced signals about the global infection level, towards the end, we mention the possibilities of having personalized signals for every individual. This thesis also includes, in the appendix, another model which captures a complementary action set of individuals (where people choose who they wish to interact with). Combining these two models to allow for individuals to choose their neighbours as well as the effort they would take while meeting these neighbours will better capture the real-world process and is an interesting open problem.