A Comparison of Two Sampling Inference Systems between Design-Based Inference and Model-Based Inference

摘要

Traditional randomization-theory-based sampling methodologies believe that the values of variables on population elements are fixed and the randomness embodies in sample selection. Its inference for the population depends on sampling design. Model-based inference thinks that population elements' values are random, and the finite population is a random sample drawn from a superpopulation. Inference for the population depends on modeling. This paper compares the two methods on application situation, weights and assumed conditions, and points out that model based inference have important application value in complex sampling.