|
Abstract:
|
Previous work regarding lowering prediction error in model-based sampling strategies have different shortcomings. For example, optimal strategies often require intensive computation, also depend on the population model and the predictor. The spatial systematic design cannot work properly under an anisotropic population, or population with nonhomogeneous variance. Intuitively, one would like to select units that account for as much population variability as possible. Ideally, it can be done by carefully examining the population covariance matrix, in which the information of the population variability is contained. The object of this study is to construct several model-based sampling designs which make use of various multivariate analysis techniques to explore the information contained in the population covariance matrix. The properties of these designs will be discussed and compared.
|
