Back to PLS Helpmultiple comparisons problem?
Posted on 02/02/07 12:46:59
Number of posts: 100
Author: Brian Luus (---.vc.shawcable.net)
Date: 01-17-07 14:57
Hi,
When presenting data for a PLS analysis of event-related MEG data, a criticism came up that the PLS algorithm makes multiple comparisons and the reason that bootstrapping is used is to try and correct for the multiple comparisons problem.
On the second column of page S252 of the 2004 Neuroimage Paper, it's explained that since no statistical test is performed, there is no multiple comparisons problem.
How could I try to convince the skeptic?
Thanks,
Brian
Date: 01-17-07 14:57
Hi,
When presenting data for a PLS analysis of event-related MEG data, a criticism came up that the PLS algorithm makes multiple comparisons and the reason that bootstrapping is used is to try and correct for the multiple comparisons problem.
On the second column of page S252 of the 2004 Neuroimage Paper, it's explained that since no statistical test is performed, there is no multiple comparisons problem.
How could I try to convince the skeptic?
Thanks,
Brian
Replies:
Re: multiple comparisons problem?
Posted on 02/02/07 12:47:20
Number of posts: 100
Author: Randy (---.rotman-baycrest.on.ca)
Date: 01-18-07 14:39
Brian Luus wrote:
>
>
> How could I try to convince the skeptic?
>
smacking them usually helps...
In all seriousness, however, it is important to keep in mind the difference between what the permutation test is for vs. bootstrap. The statistical test for the null hypothesis is done with permutation tests and is done at the level of the entire latent variable (LV). Since LV's are mutually orthogonal you need not correct for multiple comparisons here. The bootstrap estimation is not a null hypothesis comparison, but rather one of reliabililty estimation - i.e., given a significant LV, which data points in the LV provide the most reliable contribution to the effect? You can liken it to a confidence interval estimation.
Randy
Date: 01-18-07 14:39
Brian Luus wrote:
>
>
> How could I try to convince the skeptic?
>
smacking them usually helps...
In all seriousness, however, it is important to keep in mind the difference between what the permutation test is for vs. bootstrap. The statistical test for the null hypothesis is done with permutation tests and is done at the level of the entire latent variable (LV). Since LV's are mutually orthogonal you need not correct for multiple comparisons here. The bootstrap estimation is not a null hypothesis comparison, but rather one of reliabililty estimation - i.e., given a significant LV, which data points in the LV provide the most reliable contribution to the effect? You can liken it to a confidence interval estimation.
Randy
Re: multiple comparisons problem?
Posted on 02/02/07 12:47:37
Number of posts: 100
Login to reply to this topic.
