Abstract:
Adaptive Cluster Sampling (ACS) design was introduced as a powerful tool for assessing populations (clustered, rare, hidden or hard to reach such as rare plants or animal species, drug users, people engaged in illegal activities, people infected by HIV and AIDS among others) that are difficult to estimate using conventional sampling methods. The ACS is such that whenever an observed value of a unit selected satisfies a given criterion more units in the neighborhood are added to the sample and are observed. Adaptive Cluster Sampling with a data driven stopping rule (ACS') was proposed to control the final sample size and to prevent from efficiency loss when no prior knowledge of population structure is available. The usual design unbiased estimators (Horvitz-Thompson, H-T and Hansen-Hurwitz, H-H) applied on ACS' design are however not a Function of sufficient statistics and hence can be improved. This study proposed to apply the Rao-Blackwell method of conditioning on sufficient statistics and conditioning on minimal sufficient statistic on the estimators (H-H and H-T) under ACS‘design. Varied population structures were simulated. Results from simulated populations show that with the Rao-Blackwell method, ACS' design is preferable to the ordinary ACS design for highly and some less aggregated population in tem1s of efficiency and cost. The design is also more efficient than the ordinary ACS design for populations that are not rare or clustered.
