Back to PLS HelpPermuting and Bootstrapping with only 1 condition
I'm Online
Posted on 06/02/09 14:16:48
Number of posts: 8
Hi All,
I've taken Randy's advise and so find myself using commnad line PLS to perform a behavioral PLS on my cortical thickness data. As I am analyzing cortical thickness data, I find myself declaring that I am using only 1 "condition". I am curious whether this has any effect on the permutation and bootstrapping tests. Specifically:
(1) as the permutation test normally shuffles subjects between conditions within the design matrix, how can I get a valid permutation test, given that I have only 1 "condition"? When I try the PLS, the command line script reports that it *is* doing the permutation test -- and I do get believable-looking results -- but is this, in reality, OK?
(2) when I try to do the bootstrap, I get an error message .. "bootstrap order is not available". Might this be related to my 1 condition? Is so, what would be the suggested way to threshold my singular image?
Thanks,
-Jim
I've taken Randy's advise and so find myself using commnad line PLS to perform a behavioral PLS on my cortical thickness data. As I am analyzing cortical thickness data, I find myself declaring that I am using only 1 "condition". I am curious whether this has any effect on the permutation and bootstrapping tests. Specifically:
(1) as the permutation test normally shuffles subjects between conditions within the design matrix, how can I get a valid permutation test, given that I have only 1 "condition"? When I try the PLS, the command line script reports that it *is* doing the permutation test -- and I do get believable-looking results -- but is this, in reality, OK?
(2) when I try to do the bootstrap, I get an error message .. "bootstrap order is not available". Might this be related to my 1 condition? Is so, what would be the suggested way to threshold my singular image?
Thanks,
-Jim
Replies:
Permuting and Bootstrapping with only 1 condition
I'm Online
Posted on 06/05/09 13:02:28
Number of posts: 8
Hi List,
Ok, I've solved my bootstrapping problem ... which also mostly addresses my permutations question. For posterity, here's what happened ...
When running the commandline PLS, the script asks for the number of unique subjects to be included in the bootstrap, and suggests 50%. Problem is that, while it suggests 50%, it actually defaults to an empty list. This messes up processing further down the line. So, my error was in misreading the message and assuming a default value of 50%. Now, having accepted primary responsibility, might I also suggest that either (1) the default *should* be set to 50%, or (2) a more meaningful error message be presented either in pls_analysis.m, or rri_boot_check.m, or rri_boot_order.m. A meaningful error message would have saved me from myself.
BTW, it would appear as if, when specifing only one condition, the permutations *are* done within that condition (how else?).
-Jim
Ok, I've solved my bootstrapping problem ... which also mostly addresses my permutations question. For posterity, here's what happened ...
When running the commandline PLS, the script asks for the number of unique subjects to be included in the bootstrap, and suggests 50%. Problem is that, while it suggests 50%, it actually defaults to an empty list. This messes up processing further down the line. So, my error was in misreading the message and assuming a default value of 50%. Now, having accepted primary responsibility, might I also suggest that either (1) the default *should* be set to 50%, or (2) a more meaningful error message be presented either in pls_analysis.m, or rri_boot_check.m, or rri_boot_order.m. A meaningful error message would have saved me from myself.
BTW, it would appear as if, when specifing only one condition, the permutations *are* done within that condition (how else?).
-Jim
Untitled Post
Posted on 06/05/09 13:09:54
Number of posts: 394
Hi Jim,
Sorry that we have been slacking at getting back to you. I am glad you solved the problem and we will look into making sure the default value is set as such.
Regarding the permutation test for behavior vs brain when there is 1 condition, consider the permutation test as asking the question:
"If you take any random pairing of behavior and brain do you get as large a singular value as with the correct pairing?" This would apply to any number of conditions. It is not the same as the permutation for "activation" analysis where one looks only at mean signal. In that case, the permutation on a single condition would make no sense,
Randy
Sorry that we have been slacking at getting back to you. I am glad you solved the problem and we will look into making sure the default value is set as such.
Regarding the permutation test for behavior vs brain when there is 1 condition, consider the permutation test as asking the question:
"If you take any random pairing of behavior and brain do you get as large a singular value as with the correct pairing?" This would apply to any number of conditions. It is not the same as the permutation for "activation" analysis where one looks only at mean signal. In that case, the permutation on a single condition would make no sense,
Randy
Login to reply to this topic.
