BUCHI NIRCal - Manual (Page 17)

Introduction
NIRCal 5.5 Manual, Version A 17
The choice of parameters and calculation methods can be done automatically by using the Calibration
Wizard.
Figure 2: For the review of a quantitative calibration the prediction values
according to equation [1] are compared with the reference values. In an
ideal case, the corresponding points are lying on the 45° calibration curve
through the zero point both for the calibration and validation samples.
Inspection of the quantitative calibration

Precision the SEC and SEP provide the magnitude of the standard deviation for
the calibration set and the independent validation set. The two values
should be as small as possible, but they are likely to be comparable
with the standard deviation of the conventional laboratory
method. With an acceptable calibration, the two values are also
roughly equal (Consistency: »100)

Accuracy the V-Set Bias provides information on the average deviation of the
predicted values from the true values. This value gives information on
a systematic deviation of the calibration and therefore should be as
close to zero as possible. The C-Set Bias is zero by definition.

Regression
coefficients, r show how well the predicted values (NIR values) match the reference
values (original property values) on average.

Q-Value for optimisation, different parameters are integrated in this quality
factor. The Q-Value lies between 0 ( = inoperative calibration) and 1 (
= ideal calibration)
Possible reasons for bad calibration results
Outliers can be recognised for the differently calculated calibrations as big differences between
predicted values and reference values appear.
Remedy: Outliers must be erased from the calibration as well as the validation set. This always must
be combined with a careful clarification of the reason for the appearance of the Outlier.
To find out if a reference value or the measured spectra must be regarded as an Outlier, the score
plots should be reviewed (Graphics / Scores / 2D-Scatter).