两种统计检验法在异常评分筛选中的应用研究  

Research on Application of Two Statistical Tests in Screening of Abnormal Score

作　　者：刘伟力 姜箴 丁春光[1] 刘泽超 LIU Weili;JIANG Zhen;DING Chunguang;LIU Zechao(CEPREI,Guangzhou 510610,China)

机构地区：[1]工业和信息化部电子第五研究所,广东广州510610

出　　处：《电子产品可靠性与环境试验》2020年第S01期16-19,共4页Electronic Product Reliability and Environmental Testing

摘　　要：在目前的项目评审中,项目能否通过取决于专家的主观评分。为了避免其中可能存在的异常评分,传统的做法是掐头去尾保留中间值后求平均值,但这种方式简单地去掉了最高和最低分,不利于准确地反映专家的意见。将专家打分视为一组统计数据,使用统计检验方式来检验异常分数,并对比了Grubbs检验法和Dixon检验法在此应用场景下的表现,得到了在专家个数足够多的情况下使用Dixon检验法筛选异常值更为准确的结论。In the c urrent project review,whether the project can pass or not depends on subjective scoring of experts.In order to avoid possible abnormal scores,the traditional method is to remove the maximum value and minimum value and then average the remaining scores.This method simply removes the highest and lowest scores,which is not good for accurately reflecting the opinions of experts.The expert scores are considered as a set of statistical data,and the abnormal score is tested by statistical method,and the performance of Grubbs test and Dixon test in this application scenario are compared.And it is concluded that using the Dixon test to screen outliers is more accurate when the number of experts is sufficient.

关 键 词：专家评审 统计检验 离群值 格拉布斯检验法 狄克逊检验法 

分 类 号：TB114[理学—概率论与数理统计]