Back to PLS HelpPLS and p-values
Posted on 08/27/08 09:02:54
Number of posts: 29
j crofts
Hi,
I've been trying to get my head around what exactly PLS is doing and some feedback would be helpful. Is it correct that PLS (mean centered) is basically doing a principle component analysis on the average patient across groups? And that I can then interpret the resulting permutation tests and boostrap as a measure of how well each group is approximated by the average patient? For low group variation greater structure is present and the stats measures in some sense the amount of structure contained in the data. Large variance would basically amount to a random (structureless) set of data, resulting in bad stats.
Cheers,
Jonathan
I've been trying to get my head around what exactly PLS is doing and some feedback would be helpful. Is it correct that PLS (mean centered) is basically doing a principle component analysis on the average patient across groups? And that I can then interpret the resulting permutation tests and boostrap as a measure of how well each group is approximated by the average patient? For low group variation greater structure is present and the stats measures in some sense the amount of structure contained in the data. Large variance would basically amount to a random (structureless) set of data, resulting in bad stats.
Cheers,
Jonathan
Replies:
Untitled Post
Posted on 08/27/08 10:05:27
Number of posts: 394
The mean-centered PLS does a PCA on group data averaged within-task and expressed as a deviation from the group's grand mean.
The permutation test reassigns the condition labels randomly and redoes the mean-centering. The assumption here is that if the data are random, then any reassignment of condition label should produce a singular value (multivariate covariance) as large as the correct condition labels.
The bootstrap is an estimate of the reliability of the mean deviation. If subjects are quite different in the distance from the grand mean by task then the reliability will be low (bootstrap estimated standard error will be high).
Does this help?
cheers
Randy
The permutation test reassigns the condition labels randomly and redoes the mean-centering. The assumption here is that if the data are random, then any reassignment of condition label should produce a singular value (multivariate covariance) as large as the correct condition labels.
The bootstrap is an estimate of the reliability of the mean deviation. If subjects are quite different in the distance from the grand mean by task then the reliability will be low (bootstrap estimated standard error will be high).
Does this help?
cheers
Randy
Untitled Post
Posted on 08/27/08 10:46:38
Number of posts: 29
j crofts
Hi Randy,
Yes, thats very useful.
Cheers,
Jonathan
Yes, thats very useful.
Cheers,
Jonathan
Login to reply to this topic.
