I have hydro logical secondary data set. I have done standardization of the data. Now is it necessary to log transform a data set for doing PCA analysis for removing skewness?
If any of your data is proportional, it's recommended to arcsine-square root transform it.
You don't have to transform, but because PCA is modelling linear/monotonic gradients you often need to.
I'd do this first though (log then standardise).
Hi, Jay, any statistical data to obtain a comparable results before possible transformation should check for "normality" of a distribution. Note that normalization and standardization (it is actually substitution of real var by standard var. are different procedures. Your distribution graph will show a type of necessary transformation. I'm illustrating this by figure from my paper on species richness (Statistica 8). The upper fig shows distribution slightly different from "normal" p>0,05, and down fig is of normal distribution. It (upper) can be norm. by logarithms. But actually in this case differences are so small that not need any norm.
Standardization you need in case of necessity to compare absolutely different units of measurement - like abundance and biomass in Biology.
I hope it helps,
Andrey
- Badges
- Science topic