Estimating the Size of Clustered Hidden Populations

Abstract

Successive sampling population size estimation (SS-PSE) is a method used by government agencies, aid organizations, and researchers around the world to estimate the size of hidden populations using data from respondent-driven sampling surveys. SS-PSE addresses a specific need in estimation, since many countries rely on having accurate size estimates to plan and allocate finite resources to address the needs of hidden populations. However, SS-PSE relies on several assumptions, one of which requires the underlying social network of the hidden population to be fully connected. We propose two modifications to SS-PSE for estimating the size of hidden populations whose underlying social network is composed of disjoint clusters. The first method is a theoretically straightforward extension of SS-PSE, but it relies on prior information that may be difficult to obtain in practice. The second method extends the Bayesian SS-PSE model by introducing a new set of parameters that allow for clustered estimation without requiring the additional prior information. After providing theoretical justification for both novel methods, we then assess their performance using simulations and apply the Clustered SS-PSE method to a population of internally displaced persons in Bamako, Mali.