Hello
I'm doing structural PLS with four different groups using the anterior and posterior hippocampus as seeds, and I'm a bit confused about the results I'm getting.
In my analysis the 2nd and 4th latent variables come out as significant, but not the 1st. It's my understandning that the 1st LV is pretty much always significant and I'm wondering why this is not the case in my analysis? I tried to increase the permutations and bootstrap ratios (from 1000/500 to 5000/5000) but the only difference is that the 4th LV is no longer significant.
I have checked the behavioral plots for outliers but it doesn't look like thats's the problem here. When I look at the correlation overviews of the 1st LV, it does look like each group show exactly the same structural covariance patterns, and I'm thinking that maybe this LV consists of areas of covariance that are the same for each group. Hence it is not that strange that it doesn't reach significance, but when I ask colleagues who have done PLS analyses in the past, they all seem a bit concerned about the 1st LV not being significant.
So, is this a problem and if it is, what should I do?
All the best,
Eva
Eva.. can you provide some more information
1) what are the data? How many structural measures?
2) what is the design? Is this just a seedPLS using the values in the two seed regions?
thanks,
Nancy
Eva.. can you provide some more information
1) what are the data? How many structural measures?
2) what is the design? Is this just a seedPLS using the values in the two seed regions?
thanks,
Nancy
Thank you for replying, Nancy.
I hope I understand your questions correctly.
The data is VBM processed grey matter volume data. I've used two ROI's; anterior and posterior hippcoampus, and extracted the seed volumes from unsmoothed images. Then I looked at the covariance patterns of the seed volumes with rest of the brain (smoothed images).
The design is 2x2x2 (group, condition, seed region). In this particular case it´s men with genotype A, men with genotype B, women with genotype A and women with genotype B (plus the two seed regions). For each of the four groups I´ve created separate sessiondata.mat files.
Best,
Eva
Hi Eva..
thanks alot.. that helps..
Next step - can you post a copy of the "correlation overview" for each LV, and the singular values plot to a dropbox (or equivalent)? (The PLS message board cannot take attachments/figures).. (pdf or powerpoint is fine)
Nancy
Here is a link to a dropbox folder. The files are .jpg, I hope that works.
https://www.dropbox.com/sh/rjiyaf5r83ipnk1/AAChyxfAxqtMh4HJoD2H-R2Ra?dl=0
And also, I should add that the increased permutations did not result in LV4 loosing significance, that was due to me extracting seed volume data from unsmoothed images instead of smoothed (as I did in the beginning of my analyses). So, 5000 permutations resulted in the same number of significant LV´s as 1000 permutations. Sorry about this confusion.
Best,
Eva
Hi Eva - I have had some offline interactions with Hedvig on this and will quote the outcome of the exploration below.
"You have what called a "spherical solution" (or at least very near) wherein there is no dominance for a a given LV so the "eigenspace" is closer to sphere rather than an ellipse. Typically you will see LV1 have a very large singular value (with ~50% of the covariance) but if its not dominant the proportions will be more similar. This leads to the permutation outcome you observed. THis is not a problem, but a fact of the data. So, this by way of saying you can go ahead an interpret the significant LVs and ignore the others even if they are "out of order"."
If you need further clarification, please let me know.
Randy
Thank you so much for clarifying, Randy. That's great news.
//Eva
Baycrest is an academic health sciences centre fully affiliated with the University of Toronto
Privacy Statement - Disclaimer - © 1989-2024 BAYCREST HEALTH SCIENCE. ALL RIGHTS RESERVED