能力验证检测数据的两种稳健统计分析方法比较及结果合理性判断

四分位法和迭代法均是计算检测能力验证数据目标标准偏差的稳健统计分析方法, 其计算结果将直接用于判断实验室上报数据是否合格。利用195套(100多个检测项目)检测能力验证数据, 比较了两种稳健统计分析方法计算结果。模拟和实际数据计算结果均表明, 当数据分布趋势是正态分布时, 两种方法结果基本吻合, 但对于数据分布明显偏离正态分布检测项目, 四分位法计算得到的标准偏差过严, 导致实验室上报数据满意率明显下降, 出现统计学“弃真”错误。以删掉离群值后的经典统计法结果作为依据, 对两种方法合理性进行了判断。对于固体样品, 四分法与迭代法结果(标准偏差)比较接近, 两者相对偏差平均值为5.7%;液体样品为13.8%, 两者差别较大。迭代法既减小了离群值对统计分析结果的影响, 又避免了判断离群值时采用不同规则可能带来的人为因素影响。建议尽快将迭代法纳入国内检测能力验证数据评价标准体系。

Comparison of two robust statistic methods used in inspection data of proficiency testing and evaluation of its results rationality

Quartile and iteration are two robust statistic methods used to calculate target standard deviation of data of proficiency testing, whose results are directly applied to evaluate whether the laboratory data are qualified.The analysis results of the two robust statistic methods are compared based on 195 sets of data (more than 100 test items) of proficiency testing.The calculation results of both simulation and real data show that the results of two methods are basically consistent when data tend to normal distribution.But when the testing item data obviously deviates from normal distribution, the standard deviation from quartile method is so strict that the pass rate of laboratory data is obviously decreased and it appears to make statistically "abandoning true" mistake.The rationality of two methods is evaluated based on the results from classic statistics deleted outliers.The average of relative deviation between two results results is 5-7% , which means the two results are relatively closed to each other for solid samples.The relative deviation of two methods’ results is 13.8% for liquid samples, which means the difference of two methods is relatively large.Iteration method not only reduces the effect of outliers on statistical analysis results but also avoids the human factor in the identification of outliers.It is suggested that the iteration method should be applied in domestic standards system in proficiency testing as soon as possible.