BUCHI NIRCal - Manual (Page 174)

NIRCal 5.5 Software Manual
174 NIRCal 5.5 Manual, Version A
3.18.18 CV Regression Coefficients[1] t-test
Description Uncertainty testing in regression model.
Use Used for selecting most relevant and stable wavelength.
Method PCR / PLS
Matrices ID 92
Tip
Details
Related Topic
Explanation:
t = Abs( xm ) / sx * sqrt(g)
where
t = CV Regression Coefficients[1] t-Test;
xm = Regression Coefficients[1] (of the final calibration);
sx = CV Regression Coefficients[1] Sdev;
g = number of CV groups.
What are good prediction equation coefficients?

Where Abs( xm ) is big, there are wavelength with a huge relevance to the prediction
equation;

Where sx is small there are stable wavelengths in the prediction equation.
Due to this 2 points above, consider only at t because it combines the aspect of both, so deselect
wavelengths with a small t value.
To get a statistical based limit, look in a t-distribution table, select a alpha i.e. 0.05 and use a degree of
freedom of g-1.
Recommended readings:
[Uncertainty testing in PLS regression
” A.M.C.Davies, Norwich NIR Consultancy, 75 Intwood Road,
Cringleford, Norwich NR4 6AA, UK]
http://www.spectroscopyeurope.com/TD_13_2.pdf
3.18.19 CV Regression Coefficients[1] Variance to final Calib
Description Sum of squared differences of the Property Spectrum of the final calibration
to each CV CV Regression Coefficients [1] of each group.
Use Used for the CV Regression Coefficients [1] t-test.
Method PCR / PLS
Matrices ID 91
Tip
Details Sum( Sqr( B - B(g) )
Formular B8-CV SECV
Related Topic CV Regression Coefficients[1] t-test