Building estimates for totals in respondent driven sampling

Abstract

Respondent-driven sampling (RDS) is a widely used methodology for estimating characteristics of hard-to-reach populations, such as homeless individuals, undocumented immigrants, and indigenous communities. Despite its effectiveness in obtaining representative samples, RDS lacks a robust estimation framework for total population figures. This limitation hinders its application in disaggregating Sustainable Development Goal (SDG) indicators, which are crucial for monitoring marginalized groups under the “leaving no one behind” principle of the 2030 Agenda. Building on previous research, we propose an improved estimator for total population counts based on the Generalized Weight Share Method (GWSM). Our approach employs a multi-phase sampling technique and ensures unbiased estimates when post-seed selections are random. Additionally, we introduce an approximation method for cases where some network information is unknown. We compare our method to traditional RDS estimators, highlighting its advantages and limitations. The paper is structured as follows: we first analyze RDS as an indirect sampling method, then extend our model to scenarios with partial link observations. We also address potential biases and present empirical analyses that validate our approach. Our findings contribute to refining RDS estimation techniques, enhancing its reliability for policy-relevant data collection.