Advances in Adaptive Sampling

Abstract

Many populations defined by illegal or stigmatized behavior are difficult to sample using conventional survey methodology. Respondent Driven Sampling (RDS) is a participant referral process frequently employed in this context to collect a sample. It operates by incentivizing current study participants to recruit their social connections. This sampling methodology records community structure that provides invaluable information about contagions in the relevant population. However, it suffers from a complex sampling design and a rigid recruitment procedure. In this thesis, we find current RDS methods inefficient and overly reliant on simplifications of the RDS design. In response, we propose leveraging the inherent incentive structure of this sampling mechanism to achieve an explicit study goal. Given the sequential nature of RDS, online reinforcement learning (RL) is a natural framework for ascertaining an optimal strategy for incentivizing recruitment. We suggest nearly optimal methods, inference procedures and extensions for this promising new adaptive network sampling technique.