PCA with nominal, discrete and continuous data

[Kim Konings]

Hi there

Please help. I have a data sheet with 37 species (rows) and 6 traits (columns). Two of these traits are nominal, three are discrete and one is continuous. I want to group these species according to their traits. PC ORD is the only software I have found that accepts having mix-mode data for a PCA and cluster analysis. My results seemed to make sense.

However, after chatting with a statistician, he was quite surprised that I managed to do a PCA with mix-mode data, and said one can't do it with nominal or discrete data types. He said I should try to substitute my categories with different numbers and see if I get the same results. I did this and did not get the same results. This was quite disconcerting as I felt led to believe that labeling the variables as quantitative and qualitative would be enough. Do you have any solution to how I can analyse my data correctly using PC ORD or any other software or  using transformations etc? 

kind regards,

Kim

[Susan Will-Wolf]

Hello Kim,

It is my understanding that one can use qualitative/categorical variables only in the second matrix of an ordination method, including PCA, or for classification. This is the matrix used to overlay vectors or code classes of entities (species) on an ordination.

Even binary data are to be coded as quantitative in the main data matrix for an ordination or classification in PC-ORD.

If the states of a trait can be coded as an ordinal variable, that can work as a quantitative variable. Otherwise you can create a set of dummy variables to represent each trait; each dummy variable is coded as present or absent for one state/level/category of that trait. That of course means that there is no direct linkage in the data set of the suite of variables related to a single trait, if the trait has more than two states/levels/categories.

Best of luck with your analysis.

Cheers, Susan WW

Susan Will-Wolf
Senior Scientist emerita
Botany Dept., University of Wisconsin
430 Lincoln Drive, Madison, WI  53706-1381
Phone:  608-262-2754
FAX:  608-262-7509
email:  sww...@wisc.edu

---

From: pc-...@googlegroups.com <pc-...@googlegroups.com> on behalf of Kim Konings <kimko...@gmail.com>
Sent: Friday, June 26, 2015 9:06 AM
To: pc-...@googlegroups.com
Subject: PCA with nominal, discrete and continuous data

 

--
You received this message because you are subscribed to the Google Groups "PC-ORD" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pc-ord+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.