Page 25 - An Introduction to Analytical Atomic Spectrometry - L. Ebdon

Page 10

which would render the standard addition process invalid. The solutions are then aspirated and the
curve shown in Fig. 1.4 is plotted. The curve is extrapolated back until it crosses the x-axis, giving the
concentration in the unknown. A standard addition curve parallel to the calibration curve is indicative
(but not conclusive) of the absence of interference.

Accuracy of analyses should be checked using certified reference materials (CRMs). These are
materials ranging from botanical and biological to environmental and metallic samples that have been
analysed by numerous laboratories using several independent techniques. As a result, 'agreed' values of
the sample's elemental composition are produced. Therefore, by matching a CRM with the sample to be
analysed, the validity of the analysis can be verified. If the result of the analysis of the CRM agrees
closely with the certified value, the analyst can have more confidence that the sample preparation
procedure is adequate and that the results obtained for the samples are accurate. Such samples may also
be used as check samples. These should be analysed every 5-10 real samples. In this way, instrument
drift may be detected at an early stage and re-calibration performed as necessary.

Q. What are the advantages of the method of standard additions?

1.4.2 Sensitivity and Limit of Detection

The power of detection of any atomic spectrometric method of analysis is conveniently expressed as the
lower limit of detection (l.o.d) of the element of interest. The l.o.d. is derived from the smallest
measure x which can be accepted with confidence as genuine and is not suspected to be only an
accidentally high value of the blank measure. The value of x at the 99.7% confidence level (so called 3s
level) is given by

where X is the mean and s is the estimate of the standard deviation of the blank measures. The
bl
bl
deviations of a number of measurements from the mean of those measurements will show a
distribution about the mean. If that distribution is symmetrical (or to be more precise Gaussian), this
is termed a normal error curve. Hence there is always some uncertainty in any measurement. The
standard deviation is a useful parameter derived from the normal error curve. An estimate of the true
standard deviation s of a finite set of n different readings can be calculated from