|
Abstract:
|
Assessing a natural resource can use either model-based or design-based methodology. A spatial model may have to estimate the semi-variogram in order to predict the mean value of realized population. Good semi-variogram estimates require points with small separations. In contrast, design-based estimates are generally more precise with maximal separation. Design optimality can be judged by different criteria, even with the same objective, so that the "optimal" designs may be very different. Model-based assessment can be applied regardless of how samples were picked, but design-based methods can be used only with a probability sample. In some situations, e.g., litigation, model-free estimates can be advantageous. I contrast model-optimal and probability-optimal sampling. I also explore the notion that probability designs can be near-optimal judged by model-based and design-based criteria.
|
