Applying Adaptive Cluster Sampling to Study Hard-to-Reach Populations

This paper demonstrates how adaptive cluster sampling (ACS) can be used to draw a pseudo-probability sample for surveying hard-to-reach populations. Conventional sampling approaches, such as random digit dialing or standard area probability designs, are often infeasible for these populations because they are rare, spatially clustered, mobile, and poorly covered by existing sampling frames. Focusing on the Venezuelan diaspora in Peru and Colombia, we implement a stratified, two-stage pseudo-probability design in which primary sampling units (PSUs) are selected with probability proportional to size, and secondary sampling units (SSUs) expand adaptively when field teams encounter eligible Venezuelan households. Adaptive expansion follows a prespecified spiral search rule within each PSU, preserving transparent and replicable selection procedures.

We show that successful implementation of ACS hinges on extensive formative research and rigorous field protocols. Key design decisions include the size and geometry of initial SSUs, interviewer recruitment and cultural matching, and procedures for managing field constraints such as security risks, interviewer turnover, and heterogeneous urban layouts. While these conditions often necessitate operational flexibility, we illustrate how adherence to predefined selection rules can be maintained in practice. We also address inference and weighting under adaptive designs, emphasizing that standard estimators may be biased due to unequal and dependent selection probabilities. We therefore motivate design-based approaches, including Horvitz–Thompson–type estimators combined with multi-stage weighting adjustments for unequal selection, unknown eligibility, and nonresponse.

Together, these contributions provide a practical framework for researchers seeking to apply ACS while maintaining transparent selection mechanisms and population-relevant inference. Substantively, the paper illustrates how adaptive cluster sampling can meaningfully extend the reach of survey research in settings where administrative frames are incomplete and target populations are both mobile and unevenly distributed.