I have a set of around 160 samples from different sediment bodies, or stratigraphic units, which I am attempting to correlate using geochemical analysis of 25 elements for each sample, and a similarity coefficient often applied in correlating volcanic ashes. The literature gives no mention of assumptions of normality for the coefficient. It is assumed that each sample is representative of the variability in that stratigraphic unit, as representative sections were sampled while in the field.
To do these analyses, the sample is split to a smaller fraction that is loaded in to a plastic capsule, which contains the sample for the XRF irradiation. Some workers in my department who have done XRF on solid objects such as larger rock fragments, insist that 25-30 sub-samples must be analyzed per sample, So in other words, for each of my 160 samples, I would prepare 25-30 splits of the sample for XRF analysis. These ideas have propagated through the department. They are however using MDFA with its assumption of multivariate normality. Also, I am measuring sands, silts and clays, plus smaller granule gravels, of which hundreds if not thousands are visible to the XRF beam while in the capsules.
Being that there does not seem to be any assumption of normality with the Correlation Coefficient being used, is there any basis for this idea that I would need to prepare 25-30 splits of each sample in my case?
I apologize if this has been explained poorly.