126
8 Measurement Uncertainty
8.8 Random Errors
Random errors are a result of random variability of the value of the measured quantity.
An important characteristic of random errors is that the positive and negative values
of those errors are equally likely. Random errors occur as a result of many factors
that fluctuate during subsequent measurements (e.g., temperature, pressure, voltage).
The measure of the spread of the results due to the random errors is the standard
deviation of the mean for a given measurement series. Standard deviation is one of the
components—but not the only one—of the measurement uncertainty. Measurement
uncertainty is a parameter of a broadly understood result spread as an effect of the
influence of many partial random factors. In practice, it can happen that the resultant
of those factors can be small in comparison, for example, to the resolution of the
device. In such cases, the measurer will not observe any spread, which does not mean
that the measurement uncertainty equals zero.
8.9 Requirements Concerning the Uncertainty
The knowledge of the values of the uncertainties assigned to the obtained results is
essential for performing the comparison of the results between laboratories, clients
and institutions that use the measurement results. Experienced laboratories can fairly
judge their competences by the evaluation of uncertainties assigned to the provided
measurement results. An acceptable value of the measurement uncertainty should
always be assessed with a view of the specific requirements and should always be
agreed with the clients (see: target uncertainty). It should also be borne in mind
that in specific conditions even high values of uncertainty can be acceptable, and
sometimes it is necessary to conduct measurements in a way that allows very low
values to be obtained. Figure 8.2 demonstrates a comparison of results obtained in
different kinds of laboratories (calibration, expert, testing) for testing the content of
the given substance C [mg/kg] in a given matrix.
The best consistency of results was obtained for a group of calibration laboratories participating in key comparisons. It is worth noting that the measurement
uncertainties provided by those laboratories are also closest to one other. The biggest
dispersion of in the submitted results as well as their uncertainties can be observed
for testing (named ‘reserach laboratories’ on Fig. 8.2) laboratories participating in
inter-laboratory comparisons (ILCs); in between are the results from expert laboratories participating in proficiency testing (PT). This probably is a result of the fact
that those laboratories that participate in PT handle the type of samples that are used
in a given program, whereas those laboratories whose scope does not necessarily
include such samples participate in ILC.
8 Measurement Uncertainty
8.8 Random Errors
Random errors are a result of random variability of the value of the measured quantity.
An important characteristic of random errors is that the positive and negative values
of those errors are equally likely. Random errors occur as a result of many factors
that fluctuate during subsequent measurements (e.g., temperature, pressure, voltage).
The measure of the spread of the results due to the random errors is the standard
deviation of the mean for a given measurement series. Standard deviation is one of the
components—but not the only one—of the measurement uncertainty. Measurement
uncertainty is a parameter of a broadly understood result spread as an effect of the
influence of many partial random factors. In practice, it can happen that the resultant
of those factors can be small in comparison, for example, to the resolution of the
device. In such cases, the measurer will not observe any spread, which does not mean
that the measurement uncertainty equals zero.
8.9 Requirements Concerning the Uncertainty
The knowledge of the values of the uncertainties assigned to the obtained results is
essential for performing the comparison of the results between laboratories, clients
and institutions that use the measurement results. Experienced laboratories can fairly
judge their competences by the evaluation of uncertainties assigned to the provided
measurement results. An acceptable value of the measurement uncertainty should
always be assessed with a view of the specific requirements and should always be
agreed with the clients (see: target uncertainty). It should also be borne in mind
that in specific conditions even high values of uncertainty can be acceptable, and
sometimes it is necessary to conduct measurements in a way that allows very low
values to be obtained. Figure 8.2 demonstrates a comparison of results obtained in
different kinds of laboratories (calibration, expert, testing) for testing the content of
the given substance C [mg/kg] in a given matrix.
The best consistency of results was obtained for a group of calibration laboratories participating in key comparisons. It is worth noting that the measurement
uncertainties provided by those laboratories are also closest to one other. The biggest
dispersion of in the submitted results as well as their uncertainties can be observed
for testing (named ‘reserach laboratories’ on Fig. 8.2) laboratories participating in
inter-laboratory comparisons (ILCs); in between are the results from expert laboratories participating in proficiency testing (PT). This probably is a result of the fact
that those laboratories that participate in PT handle the type of samples that are used
in a given program, whereas those laboratories whose scope does not necessarily
include such samples participate in ILC.