Back to PLS HelpPosted on 05/01/15 16:57:55
Number of posts: 12
Hi everyone,
I just had a question in regards to my task related PLS analysis, although it may be more of a design issue. I have a dataset with two groups, who underwent the same scanning procedure with two "conditions", one being a task and the other being a rest block. I've been able to do the Task-PLS just fine, but as described above, it includes all 4 conditions. However, if I'm only interested in investigating the potential differences between the two groups only during the task condition, would it be okay to run PLS with the rest conditon de-selected, and essentially doing the analysis with only the two groups, and the one task condition? Sorry again if this is more of a design issue than a strict PLS question. Thank you.
Kam
Replies:
Posted on 05/02/15 07:05:32
Number of posts: 394
Yes you can do this sort of analysis in PLS. You are essentially looking at a group main effect rather than a group X task interaction (i.e., you can only know if this is a main effect or interaction by analyzing all conditions). One option to ensure interpretability is to do non-rotated PLS with the following contrasts:
[1 1 -1 -1]
[-1 1 -1 1]
[-1 1 1 -1]
Assume the first two columns are group 1, rest and task, and 3rd and 4th columns are for group 2. The first contrast is a group main effect, second is a task main effect and last is the interaction.
Posted on 05/20/15 23:40:13
Number of posts: 12
Hi Randy,
thank you for the response. Sorry to extend this question, but I was wondering that if I was using the same data set (2 groups, 2 conditions in which one is the task, and the other a rest), and I was interested in potential functional connectivity differences between the 2 groups with a ROI only during the task, if the deselect option was appropriate. From my understanding, I'd run the normal task-PLS, and then conduct the seed-PLS based on the voxel extraction of my ROI from the task-PLS results.
However, if I am only interested in the correlations of the ROI only during the task, would I deselect the rest condition (leaving only the 1 task condition) when setting up the task-PLS, and then run seed-PLS on this.
Thank you.
Posted on 05/21/15 05:52:33
Number of posts: 394
Hi Randy,
thank you for the response. Sorry to extend this question, but I was wondering that if I was using the same data set (2 groups, 2 conditions in which one is the task, and the other a rest), and I was interested in potential functional connectivity differences between the 2 groups with a ROI only during the task, if the deselect option was appropriate. From my understanding, I'd run the normal task-PLS, and then conduct the seed-PLS based on the voxel extraction of my ROI from the task-PLS results.
However, if I am only interested in the correlations of the ROI only during the task, would I deselect the rest condition (leaving only the 1 task condition) when setting up the task-PLS, and then run seed-PLS on this.
Thank you.
The difficulty is that by looking at comparing the task condition between groups, you have no way to distinguish between task-speciific group differences versus general group differences (i.e., group main effect). The coding I proposed for the 'non-rotated' PLS will give you the capacity to make such distinction by looking at the interaction term.
Posted on 05/21/15 13:46:04
Number of posts: 12
The difficulty is that by looking at comparing the task condition between groups, you have no way to distinguish between task-speciific group differences versus general group differences (i.e., group main effect). The coding I proposed for the 'non-rotated' PLS will give you the capacity to make such distinction by looking at the interaction term.
Okay, I think I got it. Thank you Randy.
I took your contrasts provided and ran a non-rotated PLS and had a question about the results (although no LV came out as significant, I wanted to ask about the interpretation).
[1 1 -1 -1] Group main effect
[-1 1 -1 1] Task main effect
[-1 1 1 -1] Interaction
The design scores graphs all match the contrasts inputted. However, when looking at the 3 LV's produced, those related to the task main effect and interaction match their counterpart design score/contrast, but not so for the group main effect.
Here, for group A, their task brain score is approximately +100 and their rest score is approximately -100. For group B, their task brain score is -200, and their rest score + 200 (all 4 had large CI's overlapping with 0, but assuming they did not). Would this indicate that along with an interaction, there is a group main effect (but no task main effect)? Or can you only infer about group main effects and not interactions from this contrast? Sorry for all these questions, but I am confused as to why the Task PLS brain scores do not match the contrast that I thought it would be "constrained" to. Thank you.
Kam
Posted on 05/23/15 12:45:05
Number of posts: 394
Okay, I think I got it. Thank you Randy.
I took your contrasts provided and ran a non-rotated PLS and had a question about the results (although no LV came out as significant, I wanted to ask about the interpretation).
[1 1 -1 -1] Group main effect
[-1 1 -1 1] Task main effect
[-1 1 1 -1] Interaction
The design scores graphs all match the contrasts inputted. However, when looking at the 3 LV's produced, those related to the task main effect and interaction match their counterpart design score/contrast, but not so for the group main effect.
Here, for group A, their task brain score is approximately +100 and their rest score is approximately -100. For group B, their task brain score is -200, and their rest score + 200 (all 4 had large CI's overlapping with 0, but assuming they did not). Would this indicate that along with an interaction, there is a group main effect (but no task main effect)? Or can you only infer about group main effects and not interactions from this contrast? Sorry for all these questions, but I am confused as to why the Task PLS brain scores do not match the contrast that I thought it would be "constrained" to. Thank you.
Kam
Can you send me a screen sht of the brain scores for each LV:
rmcintosh@research.baycrest.org
Login to reply to this topic.
