Author |
Message |
Katherine Bakeev (katherineb)
Junior Member Username: katherineb
Post Number: 6 Registered: 12-2009
| Posted on Friday, November 26, 2010 - 10:44 am: | |
Pati, The fact that the M distance is much greater suggests that the samples used for the validation are different form those used in developing the calibration model. You may want to look at an overall PCA analysis of the cal and val samples (if just by overlaying the spectra you do not see a difference) and see if there is any grouping in the scores plot. Regards, Katherine Bakeev |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 365 Registered: 9-2001
| Posted on Thursday, November 25, 2010 - 9:14 pm: | |
Pati - Since you give no information about the nature of the samples or the analyte, the range of the analyte, the values of the Standard Errors or any other diagnostic information, it's only possible ot give a very vague and general answer, which is the following: It can happen if the range of the analyte is small and the errors of the spectroscopic measurement relatively large, so that the "model" is not a model at all, but simply a bunch of small random numbers that vary the predicted value somewhat, around the average value of the analyte. In that case, no matter what the sample contains, the "model" will always predict a number close to the mean value of the analyte, regardelss of the spectral data. \o/ /_\ |
patrycja (pati)
New member Username: pati
Post Number: 2 Registered: 7-2009
| Posted on Thursday, November 25, 2010 - 3:01 pm: | |
Hello all, I have a problem with one of my models. I'm using Quant+ program and PCR algorithm. During model validation, the ResidulaRatio and total M-distance Ratio are much higher than critical limits, but at the same time the predicted values are very close to the values obtained by the reference method. Do you have any idea how is it possible? Regards, Pati |
|