Back to PLS HelpI'm Online
Posted on 07/30/15 09:22:52
Number of posts: 11
Dear PLS experts,
I've aplied behavioral PLS using EEG data and behavioral data for 3 conditions. The datamat is of size (22subjects * 3 conditions) x (10 elements/sources). The behavioral data is a matrix of (22*3)x 1. From the permutations, the 1st LV is significant. Now I get confused when looking at the bootstrapping results. When checking the correlations of brain scores with behavioral data of LV1, I obtain these values:
Laura
Replies:
Posted on 07/31/15 09:46:47
Number of posts: 394
Thanks Laura - can you post your results file somewhere so I can take a look at it? I tend to prefer the confidence intervals over the bootstrap ratios where possible, but I am not sure why you are getting two different patterns of correlations
I'm Online
Posted on 08/08/15 10:33:54
Number of posts: 11
Hello again,
I've sent the link to the result file by email. To check how stable these results are, I added few more sources. The main difference is that the results are flipped (from a positive to a negative sign), and the trends are flipped:
result.boot_result.orig_corr(:,1)
c1 c2 c3
-0.6127 -0.7580 -0.4706
Posted on 08/08/15 10:48:41
Number of posts: 394
which email address did you use? I havent received anything yet. please send to rmcintosh@research.baycrest.org
Posted on 08/08/15 11:18:23
Number of posts: 394
Hi Laura - I checked the correlations in the results file and am not sure how you are calculating the correlations that are different from orig_corr. Everything i see in the results file is okay
can you please clarify what exacty you are correlating?
I'm Online
Posted on 08/08/15 11:29:59
Number of posts: 11
Dear Randy,
From the bootstrapping results, the 2nd element has a bootstrap ratio over 3 (as checked from result.boot_result.compare_u(:,1). What I understand is that this element would show more strongly the experimental effect observed in the 1st LV. So I correlate the original data of this element (what I used as an input for the PLS) with behavior. It's in this correlation where I don't see the trend found in the 1st LV. Maybe they shouldn't be similar at all and I'm misinterpreting the results.
Cheers,
Laura
Posted on 08/08/15 11:44:27
Number of posts: 394
Dear Randy,
From the bootstrapping results, the 2nd element has a bootstrap ratio over 3 (as checked from result.boot_result.compare_u(:,1). What I understand is that this element would show more strongly the experimental effect observed in the 1st LV. So I correlate the original data of this element (what I used as an input for the PLS) with behavior. It's in this correlation where I don't see the trend found in the 1st LV. Maybe they shouldn't be similar at all and I'm misinterpreting the results.
Cheers,
Laura
The correlations in the results file are the correlations of the "brain scores" with behavior. The scores are the dot product on the U vector (brainlv) with the orginal data. If you are picking a particular time point in the original data, it may not match orig_corr b/c orig_corr is based on the entire data set (preferentially weighted by the values in U). The correlations should be similar but not necessarily identical.
From looking at the correlations for LV1, I would be cautious about interpreting the slight differences in the values as reflecting increases or decreases since the confidence intervals for the three conditions overlap. Basically I would interpret the LV as reflecting common correlations with behavior across tasks.
I hope that makes sense
Randy
I'm Online
Posted on 08/22/15 11:01:23
Number of posts: 11
Dear Randy,
So I can interpret it as correlations of the brain data with behavioral data as whole (regardless of the condition). Thank you for your help.
Cheers,
Laura
Login to reply to this topic.
