Comparing the Effectiveness of Robust Statistical Estimators of Proficiency Testing Schemes in Outlier Detection

This study investigates the effectiveness of robust estimators location and dispersion, used in proficiency testing listed ISO 13528:2015, outlier detection. The models utilize (a) kernel density plots, (b) Z-factors, (c) Monte Carlo simulations, (d) distributions derived from at most two contaminating one main Gaussian. simulation parameters cover a wide range those commonly encountered (PT) schemes, so results presented are fairly general application. We chose functional sub-optimal solution by grouping classifying model settings, resulting five matrices readily usable for selecting best estimator. Whenever half distribution each population is outside central distribution, there only optimal For all other cases, provide appropriate statistic. proposed method applies to 95.1% 144 an existing PT cement. These actual datasets indicate that Hampel estimator mean Q-method standard deviation performance statistic 86.1% cases.