Outliers were detected using the residual variance of the individual samples in terms of their residual variance in both the x- (NIR spectra) and y- (MoE) matrices. In a perfect model the residual variance would be zero. Consequently, outliers were selected as being those samples whose residual standard deviation was greater than two times the total standard deviation for all samples. Outlying samples were removed from both the calibration and validation data sets.