Back to PLS HelpPosted on 08/30/23 19:10:55
Number of posts: 16
Dear PLS experts -- we have run two non-rotated task-PLS contrastson fMRI data that has three conditions and two groups. Contrast 1 codes for a main effect of task [0.5 0 -.5; 0.5 0 -.5], and Contrast 2 codes for a group x task interaction [0.5 0 -.5; -.5 0 .5].
LV1 explains 75% cross-block covariance, and LV2 explains 25%.
A reviewer is asking if these differences in % cross-block covariance explained means that the "pattern corresponding to LV2 is less salient in the data"?
Our sense is that LV2 likely explains less cross-block covariance because of the between-groups factor, while LV1 is entirely within-subjects and so doesn't have to account for between-subjects variance. Is this reasoning on the right track?
All the best,
Reece
Replies:
Posted on 08/30/23 19:34:53
Number of posts: 394
Dear PLS experts -- we have run two non-rotated task-PLS contrastson fMRI data that has three conditions and two groups. Contrast 1 codes for a main effect of task [0.5 0 -.5; 0.5 0 -.5], and Contrast 2 codes for a group x task interaction [0.5 0 -.5; -.5 0 .5].
LV1 explains 75% cross-block covariance, and LV2 explains 25%.
A reviewer is asking if these differences in % cross-block covariance explained means that the "pattern corresponding to LV2 is less salient in the data"?
Our sense is that LV2 likely explains less cross-block covariance because of the between-groups factor, while LV1 is entirely within-subjects and so doesn't have to account for between-subjects variance. Is this reasoning on the right track?
All the best,
Reece
Hi Reece- the metric %cross-block has nothing to do with total variance but rather the total covariance. So LV2 is less strong relative to LV1, but why can't easily be connected to within vs between Ss variance per se. It has a small relative covariance value attached to it.
Having said that, its almost always the case in factorial analyses that the interaction terms are weaker, so its not too surprising. The exception is where you have complete cross-over interactions.
hope this makes sense
Randy
Login to reply to this topic.
