痕量测定中两种处理未检出值资料的统计方法比较

Comparison study on regression on order statistics and substitution method of estimation on nondetects

目的:比较用于处理包含未检出值的痕量测定数据的次序统计量回归方法与经典替换方法的估计效果。方法:应用Bootstrap方法设计模拟试验生成服从正态和非正态分布的模拟数据,分别用次序统计量回归与经典替换方法进行统计量的估计,与相应的原完整数据估计值进行比较,计算相对估计误差作为考核指标。结果:在数据服从正态和对数正态分布时,次序统计量回归效果明显优于简单替换法;资料服从非正态分布资料时,次序统计量回归方法没有明显优势。结论:次序统计量回归是处理左截尾数据更为有效和稳健的方法,但是,计算经典平均水平统计量时,需要考虑合适的正态分布假定。

Objective： the purpose of this study was to compare the performance of regression on order statistics （ROS） and substitution methods in estimation of nondetects. Methods： We designed a simulation experiment to generate censored data from normal or non--normal distribution, and then assess the performances of ROS and substitution using relative error to the estimation from the com- plete data. Results. Simulation results show that ROS method perform better obviously than substitu- tion for normal or lognormal distribution data, but not always better for non- normal data. Conclusion： Regression on order statistics is worth to adopt, but more attention should he paid to the assumption on distribution of this method.