I don't know what Nm3 are, even guessing that m3 means cubic metres. But the key is that you have different units of measurement for your variables. As that is so, -pca- on the covariance matrix is likely to be meaningless. Even applied to the correlation matrix for this kind of data, you may need to consider gross skewness and whether there are trace zeros.
Skewness implies logarithms, but zeros make that difficult!
Nick
[email protected]
[email protected]
I want to use a PCA with a dataset of 17 variables (congeners)
measured in 120 food and air samples.
The concentrations are expressed in
different units of measurement (ng/Nm3, ng/kg)
Is it correct assign to each
variable a value that corrispond to the concentration measured for the
congener, reported in relation to the sum of the 17 congener concentrations?
I
use the covariance matrix in PCA, with this output:
--------------------------------------------------------------------------
Component | Eigenvalue Difference Proportion Cumulative
-------------+------------------------------------------------------------
Comp1 | .0474947 .0380502 0.6869 0.6869
Comp2 | .00944454 .00531611 0.1366 0.8234
Comp3 | .00412843 .000944856 0.0597 0.8831
Comp4 | .00318357 .00184526 0.0460 0.9292
Comp5 | .00133831 .000309329 0.0194 0.9485
Comp6 | .00102898 .000329951 0.0149 0.9634
Comp7 | .00069903 .000190709 0.0101 0.9735
Comp8 | .000508321 .00011746 0.0074 0.9809
Comp9 | .000390861 .0000988767 0.0057 0.9865
Comp10 | .000291985 .000103161 0.0042 0.9908
Comp11 | .000188823 .0000422023 0.0027 0.9935
Comp12 | .000146621 .0000242402 0.0021 0.9956
Comp13 | .000122381 .0000341471 0.0018 0.9974
Comp14 | .0000882337 .0000235538 0.0013 0.9986
Comp15 | .0000646799 .0000359076 0.0009 0.9996
Comp16 | .0000287723 .0000287712 0.0004 1.0000
Comp17 | 1.03334e-09 . 0.0000 1.0000
--------------------------------------------------------------------------
2 principal components accounted for a total of 82% of variance in the data set
Is it correct?
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
