论文标题
R软件包采样标准：新的开发和扩展到空间采样
R package SamplingStrata: new developments and extension to Spatial Sampling
论文作者
论文摘要
R封装采样标准是在2011年开发的，是一种优化分层样品设计的工具。通过考虑采样框架中可用的分层变量以及对调查的目标估计值的精确约束来进行优化（Ballin＆Barcaroli，2014年）。在优化步骤的基础上，遗传算法探索了每个分配的可能替代分层的宇宙，即最佳分配，这是允许满足精度约束的微小总尺寸之一：最终的最佳解决方案是确保全局最小样本大小的最终最佳解决方案。使这种方法可行的一个基本要求是估计生成地层中目标变量的变异性的可能性。通常，由于目标变量值在帧中不可用，而只有代理值，预期的方差是通过对目标和代理变量之间的关系进行建模来计算的。在空间采样的情况下，重要的是，不仅要考虑总模型差异，还要考虑由空间自动相关得出的共变性。采样标准的最后一个版本使得可以考虑两个方差组成部分，从而可以利用空间自动相关，以获取更有效的样品。
The R package SamplingStrata was developed in 2011 as an instrument to optimize the design of stratified samples. The optimization is performed by considering the stratification variables available in the sampling frame, and the precision constraints on target estimates of the survey (Ballin & Barcaroli, 2014). The genetic algorithm at the basis of the optimization step explores the universe of the possible alternative stratifications determining for each of them the best allocation, that is the one of minumum total size that allows to satisfy the precision constraints: the final optimal solution is the one that ensures the global minimum sample size. One fundamental requirement to make this approach feasible is the possibility to estimate the variability of target variables in generated strata; in general, as target variable values are not available in the frame, but only proxy ones, anticipated variance is calculated by modelling the relations between target and proxy variables. In case of spatial sampling, it is important to consider not only the total model variance, but also the co-variance derived by the spatial auto-correlation. The last release of SamplingStrata enables to consider both components of variance, thus allowing to harness spatial auto-correlation in order to obtain more efficient samples.
