Statistics Meeting Book (February 20, 2019)

Pooled values in C metamer

Quantity C pC Quantity Recovery 103% 0.0112 pBias RSD(r) 2.99% 0.0130 pSD(r) RSD(L) 7.69% 0.0334 pSD(L) RSD(R) 8.24% 0.0358 pSD(R)

1. Unbiased and efficient estimates of Recovery, RSD(r), RSD(L) and RSD(R) are derivable from pC statistics, if for some reason you have a need for these statistics. 2. Method has excellent performance.

19

copyright 2018 by Least Cost Formulations, Ltd., all rights reserved.

Conclusions • Outliers should be identified only in the pC metamer, and should be removed only if assignable physical cause is given or the result is so deviant as to be unbelievably homogeneous with the other data. Note that platykurtosis is typical in all studies, so use of normal distribution based tests should not be depended upon to provide the sole reason to exclude labs or replicates. Assignable cause should be found. • The normality assumption must always be checked. • Use of the pC metamer allows comparison and pooling of estimates across concentrations and even across matrices. • Properly designed Proficiency Test studies with replications can provide full estimation of bias, repeatability and reproducibility if a range of different concentrations is achieved by using different distributed samples. • This study used replicated Proficiency Test data in an incremental collaborative way to validate a candidate method.

20

copyright 2018 by Least Cost Formulations, Ltd., all rights reserved.