have results with unacceptable accuracy, whilst laboratories 5 and 15 have
results closest to the true or accepted value. In practice, the situation is often
worse than this, with some laboratories incurring worryingly large determinate
errors that need to be identified and rectified.
Although accuracy is of prime importance in collaborative studies and pro-
ficiency testing, precision should also be monitored. The level to be expected
varies considerably with the concentration of an analyte and type of sample. A
useful guide for different levels of an analyte as measured by the coefficient of
variation (Topic B2) is exemplified in Figure 3.52 Section B – Assessment of data
10% 1% 0.1% 0.01% 10 ppm
Analyte concentration (log scale)1 ppm 0.1 ppm 0.01 ppm 1 ppb50403020100Coefficient of variation (%)Fig. 3. Guide to typical coefficients of variation as a function of analyte concentration.
Note that values of less than 1% should be attainable for major components of
a sample, whilst at ppb (parts per billion) levels, over 50% is acceptable.Computerized and automated analytical instrumentation facilitates the collec-
tion of large amounts of data and the simultaneous monitoring of numerous
experimental parameters. To maximize the useful information that can be
extracted, sophisticated multivariate chemometric techniquesare employed.
The mathematical computations involve matrix algebraand vectors, and rely
on the availability of specialized computer software. Typical applications
include the characterization of substances from profiles based on spectral,
chromatographic and other data, and quantitative analysis based on multiple
simultaneous measurements. Two important applications of multivariate
statistics are pattern recognition and multivariate modeling.Pattern recognition
Sets of measurements characterizing a sample, e.g. the position of prominent
infrared absorption bands (Topic H4), significant mass spectral fragments,
levels of particular analytes, and selected physical properties, are described as
patterns. These can be used to classify substances or to identify unknowns by
pattern matching. Figure 4shows the distribution of trace levels of copper and
manganese in twelve geological samples where three clusters are evident. JustMultivariate
statistics