方法标准验证实验数据中离群值的识别

Study on Detection of Outliers in Inter-Laboratory Collaboration Experimental Data for Validation of Analysis Method Standard

对方法标准验证实验中测量数据进行合格性审核,对于后续方法精密度计算是一个重要环节。文献中识别离群值的Grubbs法、Dixon法等经典方法有时不能满足要求。探讨了用稳健统计法识别离群值的可行性。基于2套文献数据和XRF方法标准验证实验精密度测量数据,对Grubbs法、Dixon法、Mandel h检验法、质控指标法和稳健统计法(四分位法、迭代法、合格数据范围判定法)进行了比较。结果表明:稳健统计法可有效识别离群值。但四分位法存在过度"检出"现象。综合考虑多种方法识别结果有利于提高离群值判定结论的可靠性。对于个别难以判断的情形,可借助质控指标、技术要求以及数据是否剔除对实验室间标准偏差的影响进行取舍。

It is an important for the calculation of the analysis method precision index to check the measurement data in interlaboratory collaboration experiment. The Grubbs method and Dixon method,which are used to identify the outliers,can not meet the requirements sometimes. It is proposed to detect outliers by robust statistical method. The results of several methods to detect outliers based on two literature data and XRF precisions experimental data from inter-laboratory collaboration are compared. It is shown that the qualified data range judgment method based on robust statistical method can effectively identify outliers. Quartile method sometimes exists ＂excessive ＂ identification phenomenon. The reliability of the conclusion can be improved by considering the results of several methods at same time. For some difficult situation in judgment,it is effective ways with the help of quality control indicators,technical requirements and whether the data deleted the impact on standard deviation between the laboratories.