In chemical analysis or physical testing, presence of a discordant value in a small number of data (n≧2) is checked quite often by using the permissible tolerance. There are several procedures for such checking, and even in ISO standards, plural procedures are adopted. In this paper, simulation was carried out on the five procedures mentioned below, and comparison was made on the following three aspects :
(1) Unbiasedness and dispersion of the final values;
(2) Cost involved in the application ;
(3) Simplicity in the statistical examination.
The five procedures compared are :
Procedure I ・・・・・・・・・・・・Procedure in ISO 4686;
Procedure II・・・・・・・・・・・・Sequential application of permissible tolerances;
Procedure III-1・・・・・・・・・Procedure in ISO 3271;
Procedure III-2・・・・・・・・・Procedure in ISO 3087;
Procedure IV ・・・・・・・・・・・・In JIS Z 8402-1974, clause 3. 3. 2 (1. 2. 2).
The following two models were used for data generation :
(1) Model-1 ・・・・・・・・・The second data is biased;
(2) Model-2・・・・・・・・・The variance of data is greater than the specified average variance within a laboratory.
Collectively for both of Model_<-1> and Model _<-2>, the procedure in ISO 3087 was proven to be the most recommendable procedure, the procedure in ISO 3271 being the second. However, when there is a limitation in the cost involved in the application of the procedure, the procedure of JIS Z 8402 in clause 3. 3. 2 (1. 2. 2) was proven to be the most recommendable.
抄録全体を表示