A Generalized Outlier Method with an Automated Test Protocol Applied to Proficiency Testing Schemes: Utilizing Monte...

Area of Science:

• Analytical Chemistry
• Statistical Methods
• Agricultural Science

Background:

• Outlier detection is crucial for accurate data analysis in Proficiency Testing (PT) schemes.
• Existing methods may not adequately address multiple or identical outliers, impacting payment analysis in industries like grain.
• Accurate outlier identification is vital for financial and quality control in agricultural commodity testing.

Purpose of the Study:

• To present a generalized method for outlier detection applicable to Proficiency Testing (PT) schemes.
• To enhance the accuracy and robustness of assigned values in automated testing protocols.
• To add value for PT participants by improving outlier detection and enabling root cause analysis.

Main Methods:

• Expansion of the Grubbs outlier detection method to identify up to 10 outliers (single, double, multiple).
• Utilized Monte Carlo simulations to establish critical values for datasets up to 1000 observations.
• Focused on a one-sided significance level of 1.25%, with six other levels available, for PT data analysis in the grain industry.

Main Results:

• Demonstrated challenges with extreme, identical outlier values due to rounding in PT data.
• Identified a multiple-5 outlier in a large dataset (>700 observations) from the French grain network.
• Highlighted the financial impact of small variations (0.1% protein in barley) and the effect of outlier exclusion (0.07% shift with quadruple outlier test).

Conclusions:

• The extended Grubbs test offers more precise mean value determination and outlier identification.
• Statistically distinct outliers, when confirmed, can provide significant insights into data integrity.
• Accurate testing protocols are essential, as minor variations can have substantial financial consequences in the grain industry.