ABSTRACT
This chapter explains how a model of the study variable can be used at an earlier stage to optimise probability sampling designs. It shows how a model can help in choosing between alternative sampling design types, for instance between systematic random sampling, spreading the sampling units throughout the study area, and two-stage cluster random sampling, resulting in spatial clusters of sampling units. The chapter examines how to use a model to optimise the sample size of a given sampling design type, for instance, the number of primary and secondary sampling units with two-stage cluster random sampling. It explores how a model can be used to optimise spatial strata for stratified simple random sampling. The models used for the sampling design are all geostatistical models of the spatial variation. The chapter illustrates the Bayesian approach with a case study on predicting the sampling variance of NO3-N in agricultural field Melle in Belgium.
