turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Principle Components Analysis clarification needed

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-05-2012 04:28 PM

Hello, All

Suppose my data have two variables A and B. **Total_Variance_Original_Data = Var(A) + Var(B) + 2Cov(A,B)**

Now I do Principle Components Analysis and get two new variables, say PC_1 and PC_2, with eigenvalues *λ_*1, and *λ_2*; **Total_Variance_PCA_Explained = λ_1 + λ_2**

My question is: are theses two variances equal to each other, **Total_Variance_Original_Data = ****Total_Variance_PCA_Explained ?**

Accepted Solutions

Solution

06-05-2012
04:43 PM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-05-2012 04:43 PM

Yes, if you do the analysis on the covariance matrix, that is, with the COV option (it is not the default option in PRINCOMP). You should analyse the covariance matrix only when your variables are roughly on the same scale.

PG

PG

All Replies

Solution

06-05-2012
04:43 PM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-05-2012 04:43 PM

Yes, if you do the analysis on the covariance matrix, that is, with the COV option (it is not the default option in PRINCOMP). You should analyse the covariance matrix only when your variables are roughly on the same scale.

PG

PG

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-05-2012 04:52 PM

Thank you very much for helping me.

If variables are not on the same scale (say, variable A is ~1000, and variable B is ~1), should I do normalization on each original variable (i.e. **(original_variable - mean)/standard_deviation** ) before doing PCA? or is there any other method?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-05-2012 05:02 PM

Exactly! PRINCOMP without the COV option does that normalization for you. - PG

PG

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-05-2012 05:03 PM

Thank you very much.