of the ability of the method to give a consistent negative result for known negatives
(see Section 16.1.2)- Analytical sensitivity: This is a measure of the change in response of the method to a
defined change in the quantity of analyte present. In many cases analytical sensitivity
is expressed as the slope of a linear calibration curve. - Robustness: This is a measure of the ability of the method to give a consistent result
in spite of small changes in experimental parameters such as pH, temperature and
amount of reagents added. For routine analysis, the robustness of a method is an
important practical consideration.
These performance indicators are established by the use of well-characterised test
and reference analyte samples. The order in which they are evaluated will depend on
the immediate analytical priorities, but initially the three most important may be
specificity, detection limit and analytical range. Once a method is in routine use, the
question of assuring the quality of analytical data by the implementation of quality
assessment procedures comes into play.1.4.3 Assessment of precision
After a quantitative study has been completed and an experimental value for the
amount and/or concentration of the test analyte in the test sample obtained, the
experimenter must ask the question ‘How confident can I be that my result is an
acceptable estimate of the ‘true’ value?’ (i.e. is it accurate?). An additional question
may be ‘Is the quality of my analytical data comparable with that in the published
scientific literature for the particular analytical method?’ (i.e. is it precise?). Once the
answers to such questions are known, a result that has a high probability of being
correct can be accepted and used as a basis for the design of further studies whilst a
result that is subject to unacceptable error can be rejected. Unfortunately it is not
possible to assess the precision of a single quantitative determination. Rather, it is
necessary to carry out analyses in replicate (i.e. the experiment is repeated several
times on the same sample of test analyte) and to subject the resulting data set to some
basic statistical tests.
If a particular experimental determination is repeated numerous times and a graph
constructed of the number of times a particular result occurs against its value, it is
normally bell-shaped with the results clustering symmetrically about a mean value.
This type of distribution is called aGaussianornormal distribution. In such cases the
precision of the data set is a reflection of random error. However, if the plot is skewed
to one side of the mean value, then systematic errors have not been eliminated.
Assuming that the data set is of the normal distribution type, there are three statistical
parameters that can be used to quantify precision.Standard deviation, coefficient of variation and variance – measures of precision
These three statistical terms are alternative ways of expressing the scatter of the values
within a data set about themean,x-, calculated by summing their total value and
dividing by the number of individual values. Each term has its individual merit. In all20 Basic principles