|
|
| Quality Parameters of the Validation |
| |
SEP is the standard error of the NIRS analysis. It has the same unit as the feature being examined, e.g. % in the determination of moisture content, or MJ/kg in the determination of energy content.
The SEP can be broken down into the bias (systematic error) and the SEP(C) (random error). According to a rule of thumb, the SEP should be a maximum of twice as large as the SEL.
Bias is the systematic error by which all the samples are on average incorrectly estimated.
If the validation samples are representative samples from the same basic totality as the calibration samples, and if the validation and calibration samples have been examined with the same reference method in the same laboratory in a short time span, then no significant bias should arise during the validation.
If a significant bias arises, then either:
- The calibration must be extended, if the validation samples derive from the same basic totality as the calibration samples, or
- The reference method must be checked, if the validation samples derive from the same basic totality, but were examined at a different time or in another laboratory.
SEP(C) is the random error of validation if a significant bias pertains. In that case, what has been said with regard to the SEP also applies to the SEP(C). The SEP(C) limit amounts to 1.3*SECV of the calibration equation.
Slope indicates the gradient of the regression lines, and should lie as close to 1 as possible, since otherwise, for example, high values will be underestimated.
RSQ is the determination measurement of the regression of the NIRS analysis values on the reference values of the validation samples. | |
|