ISO/TC 34/SC 005 - Izba Mleka

ISO/DIS 16297
in order to make sure that . Ssamples chosen to evaluate the relationship should beare representative to the
normal routine samples regarding these properties.
Factors influencing the relation can be bacteriological or non-bacteriological, for example.
These influences can be type of bacteria, growth phase, storage condition, sample preservation, geographic
differences, seasonal variations, species and breed of the animals from which the milk originates, method of
milking, disinfection, feeding methods or individual supplier.
6.2 Measurement protocol
6.2.1 Design
The evaluation of the alternative method as an estimate of the reference method requires a great amount of
different samples (typically 100 per log step). A minimum number of samples can be calculated according to
ISO 21187 | IDF 196:2004, Annex A.
Samples should:
a) be natural milk samples;
b) uniformly cover the whole range of interest;
c) be representative of the routine samples to be analysed especially taken into account the above
mentioned factors.
Samples are analysed in duplicate with the reference method as well as with the alternative method at the
same time or close to it.
6.3 Calculations
6.3.1 Visual check of a scatter diagram
Before any calculation is made a scatter diagram should be checked visually to obtain a first impression of the
character of the relationship and determine whether the expected relationship between the methods is
approximated.
6.3.2 Regression
To find a useful function to describe the relationship, several statistic approaches can be used which all have
their particularities and limits, but by combining the results a sufficiently exact solution can be approximated.
6.3.2.1 Regression models
After suitable (usually logarithmic) data transformation, the relationship between the alternative method and
the reference method is assumed to be linear within the range of interest. In that case the relationship can be
calculated by means of a simple linear regression model (y = abx + ba).
If the relationship is not linear, it is recommended to seek advice from a statistician. Methods to check for
linearity are specified in ISO 8196-3 | IDF 128-3, or use a regression program which gives the probability of
lack of fit or non-linearity.
If the repeatability error on for one or both methods after data transformation is still dependent on the level,
apply a weighted least-squares method in accordance with ISO 11095. In such cases, it is recommended to
seek advice from a statistician.
In the general case where regression is applied to calibration, the vertical y-axis (dependent variable) is used
for the reference method and the horizontal x-axis (independent variable) is used for the alternative method
8 © ISO 2010 – All rights reserved© ISO 2010 – All rights
reserved© ISO 2010 – All rights reserved