A comparison of design-based and model-based approaches for finite population spatial sampling and inference
The design-based and model-based approaches to frequentist statistical inference lie on fundamentally different foundations. In the design-based approach, inference depends on random sampling. In the model-based approach, inference depends on distributional assumptions. In this manuscript, we compare the approaches for finite population spatial data. We first provide relevant background for the approaches and then use a simulation study and an analysis of real mercury concentration data to numerically compare them. We find that sampling plans that incorporate spatial locations (spatially balanced samples) perform better than sampling plans ignoring spatial locations (non-spatially balanced samples), regardless of whether design-based or model-based approaches were used to analyze the data. We also find that within sampling plans, the model-based approaches often outperform design-based approaches, even for skewed data. This gap in performance is small when spatially balanced samples are used but large when non-spatially balanced samples are used.