1
$\begingroup$

I have a set of 20 chemical indicators measured in soil samples collected from one (the same) place over some time. What i would like to find out is: - To find which indicators are somehow correlated with others. I don't want to assume dependent / independent variables, just to see if some parameters seem to be correlated somehow. What methods should I use? I thought of cluster analysis - is that right choice? What other options I do have.

$\endgroup$
1
  • 2
    $\begingroup$ Soil sample data present some characteristic difficulties. First, it's almost always best to represent them as log concentrations, which will tend to have more symmetric distributions than the concentrations themselves. Second, unless all indicators are naturally occurring in detectable proportions, the dataset may exhibit substantial left censoring from the nondetect results. Your options depend on the extent to which that censoring occurs. A thread devoted to this issue appears at stats.stackexchange.com/questions/1781. $\endgroup$ – whuber Sep 30 '13 at 14:22
2
$\begingroup$

Quite a few, as a matter of fact.

You could start with a correlation matrix and visualise it with a heatmap.

Clustering definitely would be one way of analysing your data. Often, heatmaps as above are augmented by a hierarchical clustering diagram on the side.

Another way of understanding the structure would be to use PCA or ICA to reduce the dimensionality of the data, partitioning the variance in the variables into uncorrelated components.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.