用经验模型判断能力验证统计分析结果的合理性  

Evaluation of the Statistic Parameter of Rationality in Proficiency Testing by the Empirical Model

作　　者：孙海容[1] 李玉武[2] 

机构地区：[1]中国合格评定国家认可中心,北京100062 [2]国家环境分析测试中心,北京100029

出　　处：《岩矿测试》2012年第6期992-996,共5页Rock and Mineral Analysis

基　　金：中国合格评定国家认可委员会科技项目(2011CNAS11);国家重大科学仪器设备开发专项(2011YQ14014708)

摘　　要：能力验证是检测实验室识别与同行差异的一种途径,但中国合格评定国家认可委员会(CNAS)在组织能力验证时发现由于种种原因,有些实验室并不能秉承诚信的原则参加能力验证活动,导致统计分析结果"失真"。如何甄别这种情况,确保所有参加CNAS能力验证活动的实验室都获得公平的评价,这是CNAS亟待解决的问题。本文在用国内文献数据对经验模型进行修正的基础上,提出用上报数据目标标准偏差(或稳健标准偏差,sR)与基于经验模型的计算值之比(H值)是否小于0.5作为判据来审核上报数据统计分析结果的合理性。对13套能力验证数据分析表明:其中10套上报数据sR正常,3套数据存在问题。当目标标准偏差偏小,H<0.5时,建议采用同样是稳健统计方法的迭代法代替四分位法来计算目标标准偏差(sR),或用经验模型计算结果代替实验值计算Z值。通过验证表明,借助Horwitz经验模型可以判断能力验证数据统计分析结果的合理性,并能有效地识别并防止由于某种原因导致对能力验证结果评价失效的情况。Proficiency testing is a good approach to find the difference between individual laboratories in the same field. However, it was found that statistical results in the proficiency testing reports organized by CNAS were distorted since some laboratories reported the analysis data not based on their own experiments. How to identify this situation and ensure all participants to obtain a fair evaluation is a problem that needs to be urgently solved for CNAS. In order to evaluate the rationality of the statistical results, a ratio （be called as H） of target standard deviation （robust standard deviation, sR） to those calculated by the empirical model modified by experimental data collected from a large number of proficiency testing reports in the domestic environmental field as a criterion is proposed in this paper. Thirteen sets of data in recent years were used to validate this criterion. The results show that target standard deviations of ten sets of data are within the normal range and the part of robust parameters in three sets of data are much smaller compared to those calculated by this model. The causes for these situations were analyzed. It was suggested that the target standard deviation be calculated based on algorithm recommended by ISO 13528, and the Z-score be calculated by the calculated value from the model instead of the target standard deviation from experimental data when H is smaller than 0.5. The validation results showed that it is possible to evaluate the rationality of a statistical parameter in proficiency testing based on a modified Horwitz function, effectively identifying and preventing the situation in which the statistical results are distorted by some anthropogenic factors.

关 键 词：检测能力评价 经验模型 Horwitz函数表达式 

分 类 号：X132[环境科学与工程—环境科学] O242[理学—计算数学]