| Author | Message | ||
     J. R. Cuesta |
I would apreciate to receive some information about how to calculate the best RMS limit possible for a product, specially working with heterogeneous samples . What kind of things should be taken into account. How many subsamples should be use,.... I calculate it with the WIN ISI option and also with an Excell sheet Thanks J. R. Cuesta | ||
| Pierre Dardenne (Dardenne) |
Hi, It is difficult to fix a limit a priori. The limit will depend on the parameter you want to predict and on the quality of the model. The best is to create the models, test them on replicates and then fix RMS limit to get repeatability smaller (<30%) than accuracy (SEP). If there are several parameters involved, the limit will be fixed by the less repeatable model (parameter). Pierre | ||
| J. R. Cuesta |
Thanks Pierre, As I understand the RMS limit calculation when using a Model for prediction, must be done applying to the spectra the same math treatments than the ones used to develop the model (and not the raw spectra), and the result used in routine (in ISI Scan as an example)as RMS limit.Is this correct?. In the case of an acquisition phase for a future Model of a heterogeneus product, where subsamples are needed, the RMS should be calculate with the raw spectra because is the way the software ISI acquire the data .Doesn´t it?. In all the ISI softwares (Isi Scan, Win Scan,..) the RMS limit value is multiplied by 10^6. True?. J. R. Cuesta | ||
| Pierre Dardenne (Dardenne) |
It is not necessary to link model maths with RMS maths, but it seems logical. But the problem is there is only one RMS limit for routine analysis and you can have several models with different maths. Most of the time derivatives are used for modeling, then one of the derivatives can be used for RMS. Derivative is better to 'detect' heterogeneity than raw data which are more linked to the cell compaction. Pierre | ||
| J. R. Cuesta |
I read recently an article (Calibration Transfer in NIR Spectroscopy) from Pierre Dardenne (NIR News Vol. 13 No.4 August 2002) that helps me a lot to understand this matter of the RMS limit (limit for the sampling error). J. R. Cuesta |