Back to PLS Helpsame result with PLS running twice
Posted on 03/28/11 21:47:00
Number of posts: 16
Hi PLSusers,
In the second step 'run PLS analysis on event/block fmri data', would we get the same result when the PLS analysis is run more than once(i.e. with same conditions, same groups, same permutation and bootstrap number, the PLS analysis is done twice), do we get the same observed singular value, % and P-value?
I did the reaction time for behavioral PLS, I run the pls analysis for event fMRI data twice, but i got two different p-values but with the same observed singular value, and % value. The first time is significant but the second time is not. I am not sure what the reasons could be.
I hope it makes sense. Can anyone let me know?
Thank you.
bye
Replies:
Untitled Post
I'm Online
Posted on 03/29/11 11:04:27
Number of posts: 291
in order to duplicate your problem, please let me know where is your
data, and send me the link (please do not attach anything in email).
Untitled Post
I'm Online
Posted on 03/29/11 11:45:05
Number of posts: 229
depending on how many permutation tests you are running, it is quite possible for the results to change. The resampling matrix is generated every time you call the analysis.
If you have results that are that close to being significant, I would make sure that you are using a large number of permutation tests (>>500)
Having said that, the BehavPLS can be quite conservative. There is a possibility that you may have obtained permuted LVs stronger than the original due to the fact that correlations are bounded: depending on the magnitude of your original correlations, it can be quite easy to achieve stronger correlations with "noise" than was in the original data. It doesn't happen often, but we have seen this kind of behaviour in the past
nancy
If you have results that are that close to being significant, I would make sure that you are using a large number of permutation tests (>>500)
Having said that, the BehavPLS can be quite conservative. There is a possibility that you may have obtained permuted LVs stronger than the original due to the fact that correlations are bounded: depending on the magnitude of your original correlations, it can be quite easy to achieve stronger correlations with "noise" than was in the original data. It doesn't happen often, but we have seen this kind of behaviour in the past
nancy
Untitled Post
Posted on 03/29/11 11:48:16
Number of posts: 16
quote:
in order to duplicate your problem, please let me know where is your data, and send me the link (please do not attach anything in email).
Hi Jimmy,
The one that I run the analysis twice is the one with behavioral PLS with reaction time. I included 0 ms for all subjects in the behavioral data file and I guess that could be the reason why I got two different p-value but I am not sure about it. I only have one datamat file for the behvavioral data. I deleted the other one as I thought they would yield the same result. I am going to try another dataset and run it again to see whether I will get the same result. I will let you know then.
Thanks
bye
in order to duplicate your problem, please let me know where is your data, and send me the link (please do not attach anything in email).
Untitled Post
I'm Online
Posted on 03/29/11 11:51:49
Number of posts: 229
quote:
Hi Jimmy, The one that I run the analysis twice is the one with behavioral PLS with reaction time. I included 0 ms for all subjects in the behavioral data file and I guess that could be the reason why I got two different p-value but I am not sure about it. I only have one datamat file for the behvavioral data. I deleted the other one as I thought they would yield the same result. I am going to try another dataset and run it again to see whether I will get the same result. I will let you know then. Thanks bye
Hi Bingye...Hi Jimmy, The one that I run the analysis twice is the one with behavioral PLS with reaction time. I included 0 ms for all subjects in the behavioral data file and I guess that could be the reason why I got two different p-value but I am not sure about it. I only have one datamat file for the behvavioral data. I deleted the other one as I thought they would yield the same result. I am going to try another dataset and run it again to see whether I will get the same result. I will let you know then. Thanks bye
as I pointed out yesterday (http://www.rotman-baycrest.on.ca/index.php?action=view_thread&id=1111&module=bbmodule&src=%40random45c35fcb17881) , you cannot include the baseline in a BehavPLS if it is all zeros -you need variability in the data to use PLS, or any statistics for that matter
:-)
nancy
Untitled Post
Posted on 03/29/11 12:39:53
Number of posts: 16
quote:
Hi Bingye...
as I pointed out yesterday (http://www.rotman-baycrest.on.ca/index.php?action=view_thread&id=1111&module=bbmodule&src=%40random45c35fcb17881) , you cannot include the baseline in a BehavPLS if it is all zeros -you need variability in the data to use PLS, or any statistics for that matter
:-)
nancy
thank you Nancy.
Hi Bingye...
as I pointed out yesterday (http://www.rotman-baycrest.on.ca/index.php?action=view_thread&id=1111&module=bbmodule&src=%40random45c35fcb17881) , you cannot include the baseline in a BehavPLS if it is all zeros -you need variability in the data to use PLS, or any statistics for that matter
:-)
nancy
Login to reply to this topic.
