Abstract |
Sampling programs having the objective of describing specific real-world populations can utilize design protocols of probability sampling to ensure consistent estimation of the parameters of those real populations. Predictive models can often provide enhanced real population inference, but model-based methods are not required for rigorous inference, and are often unavailable. When model-based inference is planned, the probability sampling protocol is sometimes eliminated, with reliance more on the rigor provided by the model. Such an option is clearly feasible in certain circumstances. However, multipurpose monitoring programs are unlikely to provide those circumstances, and a probability sampling protocol is indispensable for population-scale sampling in all cases in which the properties of real-world populations are the program objective. The sampling design for EPA's Environmental Monitoring and Assessment Program (EMAP) is the illustrative example. |