cancel
Turn on suggestions
Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Try the Materials Informatics Toolkit, which is designed to easily handle SMILES data. This and other helpful add-ins are available in the JMP® Marketplace
Choose Language Hide Translation Bar
dgreen
dgreen
Level II

Rotation and steps for the Principal Components

Dec 13, 2018 02:54 PM (4474 views)

Hi everyone,

 

Recently I have started reading about principal component analysis. I have a big dataset (millions) which I will try to explain with 73 different variables. So  I would like to reduce the number of variables so I will use this method. 

 

I did the analysis and checked the scree plot. So I understood that I can use 8 components, I used "save principal components" by using red triangle and add this value. Now I have these results in my dataset.

 

What I could not understand (because of the various information in different sources) where I can see the components of each group. In principal components page I could view "formatted loading matrix". Is this the data I should see?(This is the one suggested in SAS report I believe).

On the other hand, a colleague advised me to do factor analysis after that and choose principal components again there. But then I should decide the number of components again. So if this is the case, how  I can view these results and adapt it to my study? And some of my variables are correlated with each other. So, which oblique rotation I can go with? How can I decide that?

 

I am really looking forward to see the answers.

Thank you

0 Kudos
2 REPLIES 2
jiancao
Staff jiancao
Staff

Re: Rotation and steps for the Principal Components

Dec 17, 2018 01:57 PM (4421 views)  |  Posted in reply to message from dgreen 12-13-2018
  1. The formatted loading matrix enables you to better understand the correlation patterns between the original variables and PCs with use the sliders to dim or suppress the loading values.
  2. (a) The # of factors recommended by JMP, as shown in the Factor Analysis launch dialog window, is based on the eigenvalue>1. Of course you can increase or decrease as needed. (b) People typically use Varimax for orthogonal rotation, and Quartimin (or Obquartimax) for oblique rotation. Do you want the rotated factors to still be uncorrelated? Or the oblique rotation leads to more interpretable factors?
dgreen
dgreen
Level II

Re: Rotation and steps for the Principal Components

Dec 25, 2018 02:43 AM (4373 views)  |  Posted in reply to message from jiancao 12-17-2018

Hi, thank you for your answer. I have tried different rotation methods and quartimin defintely made sense in my case. I would like to have more interpretable factors with oblique rotation. 

 

Thanks

0 Kudos