Adaptive Sampling Designs

Abstract

Adaptive sampling designs are sampling procedures that may depend on observed values of the population variable of interest, in contrast to conventional sampling designs. That is, the probability of selecting sample s is P(s | y) = P(s | ys), where y is the vector of population variable of interest and ys is the observed values in the sample. From the perspective of model-based sampling, it can be shown that the optimal sampling strategy under a given population model is in general an adaptive one. A model-based optimal two-phase sampling strategy, which gives lower mean square error than the optimal conventional strategy will be discussed. Although no optimal design-based sampling strategy exists, adaptive design-based sampling strategies can have advantages for some types of populations. The family of adaptive cluster sampling designs, which is a recent development in the last decade, has been widely applied to ecological and social network populations. It has certain advantages compared to the comparable conventional designs in the sense of giving lower mean square error and increasing sample yield, especially for hidden, rare or patchy populations. The basic principle and an example of adaptive cluster design will be introduced. For the purpose of improving the flexibility of the adaptive cluster sampling design, a model-assisted approach will be discussed as well.