死因监测整群抽样与不等概率抽样设计方案的比较

Selection of sentinel sites for death surveillance, using cluster or unequal probability sampling

为比较不同整群抽样设计方法的抽样误差及设计效应，评价不等概率抽样在死因监测中的应用效果。以陕西省107个县（市、区）作为抽样框架，采用等概率整群抽样和不等概率整群抽样等设计方案抽取样本，用复杂抽样方法计算不同方案样本的抽样误差和设计效应。不同的抽样方案得到不同的抽样误差估计，分层整群抽样的标准误小于完全随机整群抽样；不等概率抽样（πPS抽样）的设计效率虽略逊于等概率的完全随机整群抽样，但扩大了监测范围。结论：对于抽样框架明确的整群抽样调查数据，在统计分析时不应脱离预先设定的抽样设计方案和设计参数。死因监测采用不等概率抽样设计，能增加样本的权重，提高死亡率的地区代表性。

To compare the sampling errors from cluster or unequal probability sampling designs and to adopt the unequal probability sampling method to be used for death surveillance. Taking 107 areas from the county level in Shaanxi province as the sampling frame, a set of samples are drawn by equal probability cluster sampling and unequal probability designs methodologies. Sampling error and effect of each design are estimated according to their complex sample plans. Both the sampling errors depend on the sampling plan and the errors of equal probability in stratified cluster sampling appeares to be less than simple cluster sampling. The design effects of unequal probability stratified cluster sampling, such as r^PS design, are slightly lower than those of equal probability stratified cluster sampling, but the unequal probability stratified cluster sampling can cover a wider scope of monitoring population. Conclusions： Results from the analysis of sampling data can not be conducted without consideration of the sampling plan when the sampling frame is finite and a given sampling plan and parameters, such as sampling proportion and population weights, are assigned in advance. Unequal probability cluster sampling designs seems to be more appropriate in selecting the national death surveillance sites since more available monitoring data can be obtained and having more weight in estimating the mortality for the whole province or the municipality to be selected.