The results for each material/analytical method
combination were summarised by obtaining the means
and standard deviations by the AMC robust method
(Analytical Methods Committee, 1989), to avoid the
influence of any outlying data and thereby make the
procedure comparable with the Harmonised Protocol
method of handling collaborative trial results. In the
event there were very few suspect data. The summary
statistics are shown in Table 2. The reproducibility
standard deviation 1 was suitably low for the experiment,
and, on average, about 0.6 times that predicted by
theHorwitz function (Fig. 1). This finding justifies the
assumption that the test material was ‘sufficiently homogeneous’.
The plot also shows that the Kjeldahl method tends to provide a slightly lower average relative standard deviation (2.0%) than the Dumas method (2.7%). These statistics are practically the same as comparable values found in the initial study with pro- ficiency test data. As the analyses were duplicated, it was possible also to estimate the repeatability standard deviations, 1 and these are shown in Fig. 2. Here we see little or no difference between the relative precisions of the methods, with both methods showing an average repeatability relative standard deviation of about 0.9%.