A plot of the AAC residuals for each laboratory's validation
set predictions from the 18-factor standard models is shown in
Figure 5. Sixty-four of the 97 sample triplicates consisted of residuals
whose signs (positive or negative) were the same within
a triplicate, indicative that the NIR method was consistent among
laboratories. Removal of each laboratory's bias (Table I) in the
residuals did not alter the number of same-sign triplicates. A
X-square value of 87, determined from the hypothesis of a twochances-
in-eight probability of each sample's residual having the
same sign for all three laboratories, verified that the consistency
between laboratories was significant (P < 0.001). The reproducibility
error of this NIR model was 0.55% AAC.