I have developed an equation for olive paste, we can see the statistics of the database (number of samples, range, std. dev., mean), also the statistics of the equation, were the most importants are 1-VR and SECV (cross validation statistics), also important the number of PLS factors, and the Math treatments for the model.
This equations for fat and moisture has been added for routine analysis, and validated with a set of new samples of the new campaign 2011.
The validation statistics are:
Calibration performance is quite robust (not bias), and for the fat the Standard Error for Validation (SEP) is lower than the SECV of the calibration. For moisture is a little bit higher, but this is the normal case.
These samples and others, were added to the database and a new calibration developed. New equation statistics are almost similar to the old one. Anyway new variability has ben added of a new campaign and that is allways good.
If we want to improve SEP, we have to study if with some improvements in the sample presentation, replicates,....,improvement in the laboratory error,...,and so on, this statistic becomes lower, and at the same time the calibration stay robust.