Comparative study of robustness of statistical methods for laboratory proficiency testing

Area of Science:

• Analytical Chemistry
• Statistical Modeling
• Environmental Science

Background:

• Proficiency testing (PT) is crucial for ensuring laboratory accuracy.
• Evaluating the robustness of different data analysis methods in PT is essential.
• Understanding method performance under biased conditions is vital for reliable results.

Purpose of the Study:

• To investigate and compare the statistical properties and robustness of four data analysis methods used in proficiency testing.
• To evaluate the effectiveness of a novel population model for interpreting PT data, particularly in the presence of bias.
• To assess the performance of the median-nIQR, Q-method, Algorithm A, and Classical method under simulated conditions.

Main Methods:

• Monte Carlo simulations were employed to generate data from a population model.
• The model included a main population without bias and a sub-population with bias, both based on normal distributions.
• Statistical properties of the median-nIQR, Q-method, Algorithm A, and Classical method were analyzed.

Main Results:

• Significant differences in robustness were observed among the investigated data analysis methods.
• The proposed population model effectively interpreted an actual proficiency test for a dioxin isomer.
• The study highlights the varying sensitivity of methods to data bias.

Conclusions:

• The choice of data analysis method in proficiency testing significantly impacts robustness.
• The developed population model provides a valuable tool for assessing laboratory performance, especially when biases are present.
• Accurate interpretation of PT data, like that for dioxin isomers, relies on appropriate statistical methods and modeling.