Back to PLS HelpPosted on 06/10/15 06:01:31
Number of posts: 8
Hi everyone,
I want to ask you for help concerning interpretation of my results of structural behavioral PLSC. As an input we are using GM images from schizophrenia patients from VBM8 procedure and we have 5 behavioral tests for each patient, so we have just one condition. I am using 1000 permutations and 100 bootstrap and regular behav PLS. In result window the majority of BSR values are negative (for LV 1, other LVs are not significant), in correlation overview I got all behavioral measures positively correlating, and the plots in brain scores plot figure are all positively correlated (here I dont understand the y axis titled as "norm" - is it some mean value of brain scores for each patient?). What does this positive correlation means? It is correlation between brain score and behavioral (So if majority of saliences are negative I got lower brain score for voxels with more gm and higher brain score for lower gm)? With this results can I say that more GM imply lower score in behavioral test?
I am a bit confused so sorry about that and I will appreciate any answer,
Thanks in advance,
Matyas
Replies:
I'm Online
Posted on 06/29/15 16:25:19
Number of posts: 229
Hi
Not sure where the "norm" is coming from - I don't see anything like that when I plot brainscores x Behav - is that possibly a name of one of your behavioural variables???
In any case, I've put together a fake brain/behav data analysis - 5subj, 4 behav using white matter voxel probabilities -
Take a look at this and let me know if that helps answer your questions:
https://www.dropbox.com/s/fgorh295iqi09nx/StructuralOutputInterpretation.pdf?dl=0
nancy
Posted on 07/09/15 05:18:43
Number of posts: 8
Thank you very much, I think it is all clear now. I also ran some PLS analyses with only one behavioral variable to see if there some correlation between amount of gray matter and the behav. variable. Everytime I ran it I got p-values for permutation test greater than 0.5. So I shouldn´t use the results from these at all. Am I right? Just asking because I am not sure if the PLS is suitable for just one behav. variable.
Thank you in advance,
Matyas
Posted on 07/09/15 06:26:31
Number of posts: 394
with one behavior the PLS permutation test tends to be conservative - for that I would focus more on the bootstrap results and split-half. However, the full analysis where you have all the behaviors is probably much more informative and powerful.
Randy
Login to reply to this topic.
