Back to PLS Helpcorrelation or covariance?
Posted on 02/02/07 12:41:33
Number of posts: 100
Author: Zara (---.lns6-c12.dsl.pol.co.uk)
Date: 01-20-07 13:07
Hi,
I'm a bit confused about which types of PLS analyses use the cross-block covariance matrix and which use the correlation matrix in your gui?
I noticed that Lobaugh et al 2001 analysed the covariance rather than the correlation matrix in order to retain ERP amplitude information, does this mean that all PLS GUI analyses of ERP data use the covariance matrix by default, or is it different depending on which method you choose (e.g. mean-centred vs. non-rotated)?
Does PLS on fMRI data use the correlation or covariance matrix? What is the difference in outcome depending on which measure is used?
Thanks very much,
Zara
Date: 01-20-07 13:07
Hi,
I'm a bit confused about which types of PLS analyses use the cross-block covariance matrix and which use the correlation matrix in your gui?
I noticed that Lobaugh et al 2001 analysed the covariance rather than the correlation matrix in order to retain ERP amplitude information, does this mean that all PLS GUI analyses of ERP data use the covariance matrix by default, or is it different depending on which method you choose (e.g. mean-centred vs. non-rotated)?
Does PLS on fMRI data use the correlation or covariance matrix? What is the difference in outcome depending on which measure is used?
Thanks very much,
Zara
Replies:
Re: correlation or covariance?
Posted on 02/02/07 12:41:53
Number of posts: 100
Author: Randy (---.rotman-baycrest.on.ca)
Date: 01-21-07 12:54
for task PLS - mean centred or non-rotated, we always use covariances, regardless of modality. For behavior PLS, we use correlations. This is primarily b/c scaling effects in brain-behavior correlations tend to skew the SVD, so putting everything on the same scale (zscores) reduces this.
Zara wrote:
>
> Does PLS on fMRI data use the correlation or covariance matrix?
> What is the difference in outcome depending on which measure is
> used?
>
>
> Thanks very much,
>
> Zara
>
Date: 01-21-07 12:54
for task PLS - mean centred or non-rotated, we always use covariances, regardless of modality. For behavior PLS, we use correlations. This is primarily b/c scaling effects in brain-behavior correlations tend to skew the SVD, so putting everything on the same scale (zscores) reduces this.
Zara wrote:
>
> Does PLS on fMRI data use the correlation or covariance matrix?
> What is the difference in outcome depending on which measure is
> used?
>
>
> Thanks very much,
>
> Zara
>
Procrustean rotation
Posted on 02/02/07 12:42:10
Number of posts: 100
Author: Zara (---.lns4-c13.telh.dsl.pol.co.uk)
Date: 01-22-07 04:57
Hi Randy,
thanks very much for your reply!
I also have another question: is procrustean rotation by default performed during permutation and bootstrap when using the mean-centred method? In the papers (e.g. McIntosh & Lobaugh, 2004) it sounds like this is an optional step, so I just want to make sure I know what I have done to the data!
Thanks a lot,
Zara
Date: 01-22-07 04:57
Hi Randy,
thanks very much for your reply!
I also have another question: is procrustean rotation by default performed during permutation and bootstrap when using the mean-centred method? In the papers (e.g. McIntosh & Lobaugh, 2004) it sounds like this is an optional step, so I just want to make sure I know what I have done to the data!
Thanks a lot,
Zara
Re: Procrustean rotation
Posted on 02/02/07 12:42:27
Number of posts: 100
Author: Randy (---.rotman-baycrest.on.ca)
Date: 01-22-07 10:11
Hi Zara
Procrustes rotation is done for any PLS procedure that uses SVD:
mean-centred PLS
behavior PLS
multiblock PLS
It is only performed on the resampled data to ensure the SVD solutions are in the same effect space as the original data. Its not an optional step in our current implementation of PLS.
Date: 01-22-07 10:11
Hi Zara
Procrustes rotation is done for any PLS procedure that uses SVD:
mean-centred PLS
behavior PLS
multiblock PLS
It is only performed on the resampled data to ensure the SVD solutions are in the same effect space as the original data. Its not an optional step in our current implementation of PLS.
Login to reply to this topic.
