BUCHI NIRCal - Manual (Page 186)

NIRCal 5.5 Software Manual
186 NIRCal 5.5 Manual, Version A
3.18.48 PCR B-Matrix
Description Shows the influence of the PCs to the property value.
Use Useful for secondary PC selection.
Method PCR
Matrices ID 7
Tip Select the PCs with high absolute correlation value to optimize the
prediction.
Details Also known as correlation coefficient or B-matrix.
Depends only from the number of primary factors. The selected secondary
PCs take no effect on the PCR B-matrix.
Related Topic Scores , Original Property
The PCR algorithm makes as first step a Principal Component Analysis and the second step is a
Multiple Linear Regression. In the MLR the scores are multiplied with the correlation coefficients: B-
values.
Formula of MLR by PCR:
Property value = ymean + b1 * v1 + b2 * v2 + b3 * v3 + ....
where: b : correlation coefficient; v : score; 1-2: number of PC.
Here the 3. PC has very small B-value, so only the first 2 PCs are important for the parameter.
NOTE
The B-values are not normalized between -1 and + 1 as usual for correlation coefficients, because
the scores are already normalized using the Mahalanobis distance.
3.18.49 Predicted Property
Description The predicted property values for all spectra in the project with the activated
calibration. (Estimate of dependent variable)
Use Main result of a calibration.
Method all
Matrices ID 10
Tip The original and predicted property values should be as similar as possible.
Details The prediction depends on the selected secondary PCs.
Related Topic Property Residuum , Original Property
In cluster method the spectra get a 1 (one) to a class, if it is identified as such (distance and residual
limits are fulfilled).
In quantitative calibrations the predicted property values are the results of the NIR methods.