Post-Stratification against Bias in Sampling
针对存在偏差（如覆盖不足、无应答）的抽样模型，证明事后分层估计量具有极大似然性和最小方差性质，并给出偏差与方差界值，用于构建近似置信区间。
Summary For a quite general sampling model, allowing bias (due to undercoverage, nonresponse, for example), the post-stratified estimator of the population mean is shown to be maximum likelihood and have a minimal variance property. Bounds are calculated for bias and variance. In an illustration it is shown how these bounds may be used to obtain approximate confidence intervals.