Sawitzki and Golay
Method for the calculation of derivation and smoothing. In this situation, in addition to the degree of deviation (first or second), the range over which this calculation is carried out about each point is indicated (uneven number of support points, greater than 3).
SD (standard deviation)
Value for the scatter of the values of a population. The standard deviation of the samples in the validation for the NIRS analysis values should be as far as possible of the same size as the reference values. If it is perceptibly smaller than the standard deviation of the reference values, then all the samples are distorted to the mean of the analysis; in other words, the analysis values are all similar. Differences between the samples are wiped away.
with i = 1 ... n samples
SEC (standard error of calibration)
The SEC is calculated as the standard deviation of all NIRS analysis values from the reference values of the calibration samples. It is not suitable alone for the quality of the calibration to be evaluated, since under certain circumstances interference noises are also modelled during the calibration development and reduce the SEC (problem of \dq overfitting\dq).
The limit value of the SEC during calibration is dependent, among other things, on the SEL and the scatter of the features in the population. Good values are one to two times the SEL. The SEC will not be smaller than the SEL.
with i = 1 ... n samples and p wavelengths or factors in the calibration
SECV (standard error of cross-validation)
During calibration with methods of multivariant regression (PCR, PLS, MPLS), a cross-validation is carried out. In this situation, in each case a part of the samples (between 10 and 50\% depending on the size of the sample set -- 10\% at n=30, 50\% at n=200) is estimated from a calibration on the basis of the remaining samples. The cross-validation is carried out in order to identify and prevent an overfitting (see SEC) which may be being expressed in a repeat rise of the SECV.
mit i = 1 ... n cross-validation samples
SED, SEL (standard error of differences, ... of laboratory analysis)
Like SEP, except that a comparison is made not between laboratory and NIRS analysis values, but between two repeated measurements (e.g. two NIRS analyses), or two laboratory analysis (=SEL).
with 1 = 1 ... n samples and j = 1 ... r measurement repetitions of a sample.
The limit value for the SEL depends on the feature being examined, the method used, and also on the material being examined. It is to be found mostly with other designations and calculation formulae in the method books (e.g. repeatability standard deviation). Calculated from the maximum deviation of two repetitions referred to in the method specifications is the repeatability standard deviation SEL, according to:
segment
See Norris derivation
SEP (standard error of prediction)
The standard deviation of all NIRS analysis values from the reference values for the validation samples is an important measure of the quality of the NIRS analysis. It encompasses the systematic error (bias) and random error. If no bias is present (see bias), then the SEP indicates the quality of the NIRS analysis, and otherwise of the SEP(C).
with i = 1 ... samples
Example: Regarding the keyword SEP
| Number | Reference value | NIRS-analysis value | Difference | DifferenceČ |
| 1 | 100 | 101 | +1 | 1 |
| 2 | 105 | 107 | +2 | 4 |
| 3 | 110 | 110 | 0 | 0 |
| SEP | - | - | - |  |
SEP(C) (standard error of prediction corrected for bias)
The SEP(C), calculated as standard deviation of all NIRS analysis values from the reference values for the validation samples, after the NIRS analysis values have been adjusted by the bias. The SEP(C) indicates the random error. It indicates the error of the NIRS analysis in the event of a bias pertaining, which is corrected. As a limit value for the SEP(C), use is made in the software of 1.2 times the SECV. This applies to calibration sets of more than 100 samples.
with i = 1 ... n samples
Example: Regarding the keyword SEP(C)
| Number | Laboratory value | NIRS- analysis value | NIRS-analysis value corrected | Difference | DifferenceČ |
1 | 100 | 101 | 100 | 0 | 0 |
2 | 105 | 107 | 106 | +1 | 1 |
3 | 110 | 110 | 109 | -1 | 1 |
SEP(C) | - | - | - | - | |
SEV (standard error of validation)
See SEP
SG
See Sawitzki and Golay
Slope
The slope relates to the regression lines of the NIRS analysis values to the reference values of the validation samples. The slope of the regression lines should be close to 1.0. With greater deviations than 1.0, it is in particular the samples with high and low values which are being incorrectly estimated.
SNV (signal normal variance)
Important method for the correction of the spectra for influences of different particle sizes. As a rule, together with detrend as SNV and detrend is used.
t-value
The t-value is a value of the difference between a measured value and a measured value from several measured values. In spectroscopy close to IR, it is used to determine \dq outlier \dq\
A t-outlier exhibits strikingly large deviations between NIRS analysis value and reference value.
For the calibration, the t-value of a sample is calculated as
For the validation there applies accordingly
t-values greater than 2.5-3 are regarded as an indication that this sample is an outlier. These samples should be examined again on the NIRS device and in the laboratory.
Validation
For the validation, an independent random sample is taken from the same basis totality as the calibration samples. These validity samples are analyzed with the aid of the NIRS method and then examined in the laboratory with the reference method to determine their content substances. The contrast between the reference values and the NIRS analysis values is calculated statistically, in order to be able to make a statement about the precision of the NIRS analysis.
 Top |  back
 A-L
M-R
S-Z |