Dear PLS experts,
I'm running a behavioral PLS analysis, including average cortical thickness in 4 clusters and 6 physiological variables of 35 participants. Clusters were derived from a GLM comparing a patient and a control group; data were collapsed across both groups for PLS. Please see my script in the bottom.
The PLS analysis yielded 1 significant LV (“sprob”), accounting for 99% of covariance (“s.^2/sum(s.^2”) between thickness and physio variables. All four clusters are reliably associated with the LV (“compare_u”) and 4/6 physio variables are reliably (?!) correlated with the brainscores (“orig_corr”) as indicted by confidence intervals not including 0 (“ulcorr” and “llcorr”). Are these interpretations correct?
However, I am interested in the effects of physio variables on thickness, not thickness effects on physio variables, which I believe my current analysis represents. Should/could I switch the physio with the brain data (i.e., enter physio as 'datamat' and thickness as 'physiodata' in script below)? Or should I stick to my analysis but interpret other results?
Thank you very much in advance!
Nic
--------------------------------------------------
MATLAB 2011b and PLS version 6.15.1
Data files
CT_data.txt: average CT for each of 35 participants (i.e., 35 rows) in the clusters (i.e., four columns).
Physio_data.txt: values for six physiological variables (i.e., six columns) for the participant (i.e., 35 rows).
Analysis
nsubjects=35;
ncolumns=4;
in1=fopen('CT_data.txt');
datamat=fscanf(in1,'%f',[ncolumns,inf])';
fclose(in1);
save 'datamat.mat' 'datamat'
ncolumns=6;
in2=fopen('Physio_data.txt');
physiodata=fscanf(in2,'%f',[ncolumns,inf])';
fclose(in2);
save 'physiodata.mat' 'physiodata'
opt.method=3;
opt.num_perm=1000;
opt.num_boot=1000;
opt.stacked_behavdata=physiodata;
results=pls_analysis({datamat},[nsubjects], 1, opt);
save 'results.mat' 'results'
Results
# Bootstrap ratios, representing reliability of cortical thickness clusters contributing to LV
results.boot_result.compare_u
# Statistical significance of latent variables
results.perm_result.sprob
# Percentage that LVs account for covariance between thickness and physio data
pct=(results.s.^2/sum(results.s.^2))
Figures
# Output used for barplots representing correlation between brain scores and physio data along with confidence intervals:
...
m=results.boot_result.orig_corr(:,LV)
El=results.boot_result.orig_corr(:,LV)-results.boot_result.llcorr(:,LV)
Eu=results.boot_result.ulcorr(:,LV)-results.boot_result.orig_corr(:,LV)
…
# Data in "results.usc" are plotted against physio data as another way to represent the correlation between physio data and brainscores (i.e., representations of cortical thickness)
thanks for the question Nic. The model that is tested in PLS does not distinguish brain->behavior or behavior->brain so your interpretation would be the same regardless of whether you swapped the variable orders. Unless you have some way of partialing variance (e.g., mediation analysis) all you can really conclude is that there is a reliable relation between your physiological variables and cortical thickness.
hope that makes sense
Hi Randy,
Thank you so much for your very quick answer! This is indeed what I'm reporting.
I have a follow-up question - are there explicit BSR for physio data (perhaps compare_v?!)? Basically, I'm getting hung up on design vs brain LVs. I'm discussing the reliability of the clusters contributing to the LV with BSR (compare_u) and interpret the reliability of the physio data with the CIs (i.e., ulcorr and llcorr) for orig_corr. Adapted from the manual (p 69): the design LV represents how well physio variables relate to a given LV (the angle in which I'm interested) and the brain LV represents how well each cluster relates to the overall pattern exhibited by a given LV. I assume overall=relationship between LV and both, brain and physio data and is therefore reported...? (though in the figure on p 99, it says that the behavioral LV relates to the overall pattern)
Thank you very much and have a lovely day,
Nic
Baycrest is an academic health sciences centre fully affiliated with the University of Toronto
Privacy Statement - Disclaimer - © 1989-2025 BAYCREST HEALTH SCIENCE. ALL RIGHTS RESERVED