help > Principal component decomposition question
Showing 1-2 of 2 posts
Apr 12, 2017 04:04 PM | Jenna Traynor - McMaster University
Principal component decomposition question
Hi Alfonso,
I have a question regarding how to do a PCA decomposition on 6 covariates in my analysis. To give you some background, this was my original question:
I am running a within-group bivariate correlation analysis looking at the association between functional connectivity and 6 different behavioural scores, using Conn V.17.My experimental question is: is there an association between FC and each of my behavioural scores?
When I set up the findings to look for any effect among my behavioural variables 1-6, I select my patient group and the six variables and set up the contrast as so: [0 1 0 0 0 0 0; 0 0 1 0 0 0 0; 0 0 0 1 0 0 0; 0 0 0 0 1 0 0; 0 0 0 0 0 1 0; 0 0 0 0 0 0 1]. I then select all of my seeds and enter into results explorer, which shows that there is no significant effect at all.
However, if while in the second level results window, I select the simple main effect of each of my behavioural scores separately i.e. [0 1], and all of my seeds, and then enter into results explorer, I get an abundance of significant values for each behavioural score.
You suggested that entering my 6 covariates as well as all of my ROIs into the analysis simultaneously was resulting in a severely under powered analysis. You suggested to first do a PCA decomposition of my 6 covariates and then enter only the first few component scores into my second level analysis in order to increase the power of my study. However, I am unsure how to do this on my 6 covariates. I understand that you can do this on ROIs in the setup option (i.e., extract principle decomposition), but how would I do this on my covariates?
Would I extract a principal decomposition from my ROIs first in the set up option? And if so, how many PC's do I want from each ROI? Or would I do this a completely different way and do a group-PCA on all my subjects, and then go back and use only the most significant components in my analysis?
Thank you as always for your help!
Jenna
I have a question regarding how to do a PCA decomposition on 6 covariates in my analysis. To give you some background, this was my original question:
I am running a within-group bivariate correlation analysis looking at the association between functional connectivity and 6 different behavioural scores, using Conn V.17.My experimental question is: is there an association between FC and each of my behavioural scores?
When I set up the findings to look for any effect among my behavioural variables 1-6, I select my patient group and the six variables and set up the contrast as so: [0 1 0 0 0 0 0; 0 0 1 0 0 0 0; 0 0 0 1 0 0 0; 0 0 0 0 1 0 0; 0 0 0 0 0 1 0; 0 0 0 0 0 0 1]. I then select all of my seeds and enter into results explorer, which shows that there is no significant effect at all.
However, if while in the second level results window, I select the simple main effect of each of my behavioural scores separately i.e. [0 1], and all of my seeds, and then enter into results explorer, I get an abundance of significant values for each behavioural score.
You suggested that entering my 6 covariates as well as all of my ROIs into the analysis simultaneously was resulting in a severely under powered analysis. You suggested to first do a PCA decomposition of my 6 covariates and then enter only the first few component scores into my second level analysis in order to increase the power of my study. However, I am unsure how to do this on my 6 covariates. I understand that you can do this on ROIs in the setup option (i.e., extract principle decomposition), but how would I do this on my covariates?
Would I extract a principal decomposition from my ROIs first in the set up option? And if so, how many PC's do I want from each ROI? Or would I do this a completely different way and do a group-PCA on all my subjects, and then go back and use only the most significant components in my analysis?
Thank you as always for your help!
Jenna
Apr 19, 2017 12:04 PM | Jenna Traynor - McMaster University
RE: Principal component decomposition question
Hi Alfonso,
Do you have any suggestions about which to choose? Your help is greatly appreciated!
Thank you,
Jenna
Do you have any suggestions about which to choose? Your help is greatly appreciated!
Thank you,
Jenna