On Fri, 17 Jul 2015, Aaron Mackey wrote:
> One immediate question is how independent you believe the 46 components to
> be, and whether certain components could be reduced or otherwise
> coordinately-modeled; a heatmap of your 46x46 pairwise correlations should
> be informative. Also consider log-scaling, especially if some component
> fractions can be very small/minor components compared to large/major
> components.
Aaron,
Most components are elements spread across the periodic table; the rest
are composits such as total dissolved/suspended solids, acid neutralizing
capacity, specific conductance, bicarbonate.
A heat map will be drawn. Log scaling is the key to working with
compositional data. The log-ratios can be calculated using three equations
with two being most frequently applied, each depending on the analysis to
follow.
Thanks,
Rich