Analytical Sciences

Comparative Study of Robustness of Statistical Methods for Laboratory Proficiency Testing

Isao KOJIMA* and Kazutoshi KAKITA**

*Japan Accreditation Board, 1-22-1 Higashi-gotanda, Shinagawa, Tokyo 141-0022, Japan
**Nippon Steel & Sumikin Technology Co., Ltd., KSP-A101, 3-2-1 Sakado, Takatsu, Kawasaki 213-0012, Japan

A Monte Carlo approach has been applied to investigate the statistical properties of data analysis methods for proficiency testing, namely, the median-nIQR, Q-method, Algorithm A and Classical method. Simulations were carried out by using a population consisting of two normal distributions, a main population without bias and a sub-population with bias. The difference for robustness among the methods has been discussed. An actual PT for a dioxin isomer was well interpreted by using the population model employed in this study.

J-STAGE: View this article in J-STAGE