Roger Breton wrote:
For characterizing the deviation, I think the best statistics remain standard deviation. But, again, I suspect you need at least n=30 measurements to make meaningful statistical inferences from the data.
I found some references to the British Standard 597 ("Guide for the determination and reproducibility for a standard test method") in <http://www-users.york.ac.uk/~mb55/meas/ba.htm>. Unfortunately it only covers one dimensional data and two repeated measurements:
"The best way to examine repeatability is to take repeated measurements on a series of subjects. The table shows paired data [...] We expect 95% of differences to be less than two standard deviations. This is the definition of a repeatability coefficient adopted by the British Standards Institution. [...] If we have more than two repeated measurements the calculations are more complex. We plot the standard deviation of the several measurements for that subject against their mean and then use one-way analysis of variance, [...] which is beyond the scope of this article."
Klaus
