Dear PLS experts:
I need some confirmation about PLScmd: I have a dataset of 75 ROIs (SPECT), 2 groups (n = 39 and 10), I want to run mean-centred PLS off the cmd line to find the set of regions that will maximally differentiate the groups.
So:
Datamat_lst = ([39x75] [10x75]), with the ROI values
Num_cond=1
Num_subj_lst = [39 10]
Option.num_perm=1000
Option.num_boot=500
Based on my understanding of the help pls_analysis file, in the result variable:
u = saliences for brain data matrix
usc=calculated brainscores for each subject, based on saliences and ROI values
boot.compare_u = ROI’s BSR values, since you feed that into the plotroi program
perm_result.sprob = the number of times the computed singular values exceeded the observed singular values, divided by the number of permutations. These are p-values for the significance of the LVs (i.e. if the cell values are only 0.377 or 0.19, they are not significant).
Many thanks,
Kris
I can answer your first question (Question a.).
Yes, you got most of them right. For the last one (perm_result.sprob), yes, it is the number of times the computed singular values exceeded the observed singular values, divided by the number of permutations. However, since they are random permutations, the computed singlar value should be very low. If not, it means that they are not significant. I am not able to quantify it with a number, but compared with 0.91, 0.19 should be more significant.
Hi Kris...
Jimmy's correct, p=0.19 is better than 0.38, but it's not significant.
the 2nd LV is the grand mean, you can verify this by looking at the designLV (v) and is not expected to be different (usually it's much higher than what you seem to find),
Given the non-significant LV here, you would not normally use the scores in any additional analyses. You can, however, enter your behavioural measures into a two-group behavPLS. This would identify which, if any, of your behav measures have similar and/or different relations across groups with the spect measures.
cheers,
nancy
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