Rotman > Resources > Support + Services
Back to PLS Help

Mean-Centering CI vs. Non-Rotated CI
bahri
Posted on 01/03/18 10:53:37
Number of posts: 4
bahri posts:

Dear All,

After running Mean-Centering PLS we identified a significant LV (55.42% crossblock, p < 0.000, CI bars do not cross zero) then we tried to reproduce the same result with the Non-rotated Task PLS with a contrast reflecting the design obtained with mean-centering. The Non-Rotated results provided an LV with 49.81% Crossblock, p < 0.000 but with CI bars crossing zero (see attached file). Any idea on why this could be happening?

Many thanks in advance,

Best regards,
MohamedLogin to reply to this topic.

Replies:

Untitled Post
rmcintosh
Posted on 01/03/18 15:34:35
Number of posts: 394
rmcintosh replies:

Yes, its an issue with mean-centreing in the non-rotated analysis.  Can you run it without mean-centering?



Untitled Post
bahri
Posted on 01/04/18 03:34:58
Number of posts: 4
bahri replies:

quote:

Yes, its an issue with mean-centreing in the non-rotated analysis.  Can you run it without mean-centering?

Many thanks for your answer!

which option for that 0,1,2,3?



Untitled Post
rmcintosh
Posted on 01/04/18 05:43:45
Number of posts: 394
rmcintosh replies:

quote:

Many thanks for your answer!

which option for that 0,1,2,3?

Sorry, I misrembered the options - there isn't one for no mean correction (you'd think I would know that).  I thought we fixed this in the latest version to do non-rotated PLS with no mean centering.  

Can you repost the screen shot of the non-rotated result?  That didnt make it on your last posting.  Alternatively, send me an email: rmcintosh@research.baycrest.org and I will post my reply here.

Randy



Untitled Post
bahri
Posted on 01/08/18 11:16:11
Number of posts: 4
bahri replies:

quote:

Sorry, I misrembered the options - there isn't one for no mean correction (you'd think I would know that).  I thought we fixed this in the latest version to do non-rotated PLS with no mean centering.  

Can you repost the screen shot of the non-rotated result?  That didnt make it on your last posting.  Alternatively, send me an email: rmcintosh@research.baycrest.org and I will post my reply here.

Randy

Dear Dr. Randy,
 
Thank you very much for your answer and help.
Attached are the screen shot of the non-rotated result. We used the default value (0) for the mean-centering option under the non-rotated.
 
Best regards,
 

Mohamed



Untitled Post
bahri
Posted on 01/24/18 08:26:40
Number of posts: 4
bahri replies:

Dear Dr. Randy,
 
Based on some PLS papers (Edgington 1980, Good 2000 and Mcintosh 1996) I have used for one fMRI PLS analysis 600 permutations and 150 bootstraps but one of the reviewers of the paper seems to be not happy with these values and ask me to redo the analysis with at least 1000, is there a formula to calculate the needed permutation and bootstraps values? similarly, the same reviewer argues that the bootstrap threshold for neuroscience should be set to -3 to 3 (i.e. p<0.001) while we used a threshold of -1.97 to 1.97 (i.e. 0.05). In my opinion the 0.05 is correct but I would be very happy to have your opinion.
 
Thank you very much in advance,
 
Kind regards,
 

Mohamed



Untitled Post
rmcintosh
Posted on 02/03/18 07:33:04
Number of posts: 394
rmcintosh replies:

quote:
Dear Dr. Randy,
 
Based on some PLS papers (Edgington 1980, Good 2000 and Mcintosh 1996) I have used for one fMRI PLS analysis 600 permutations and 150 bootstraps but one of the reviewers of the paper seems to be not happy with these values and ask me to redo the analysis with at least 1000, is there a formula to calculate the needed permutation and bootstraps values? similarly, the same reviewer argues that the bootstrap threshold for neuroscience should be set to -3 to 3 (i.e. p<0.001) while we used a threshold of -1.97 to 1.97 (i.e. 0.05). In my opinion the 0.05 is correct but I would be very happy to have your opinion.
 
Thank you very much in advance,
 
Kind regards,
 

Mohamed

hi there

 

On the permutation and bootstrap issue, more iterations are always better and since computing power is cheap its probably better to rerun with more permutations than argue.  It will likely not make a difference.  As for the threshold, the bootstrap is not meant to be a null hypothesis test but one of reliability.  We made a point of this in the McIntosh & Lobaugh 2004 review paper.




Login to reply to this topic.

Contact Us:

3560 Bathurst Street
Toronto, Ontario
Canada M6A 2E1
Phone: (416) 785-2500

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