Back to PLS HelpOne group One behavioural measure
Posted on 02/02/07 12:17:21
Number of posts: 100
Author: sophie (---.psychol.cam.ac.uk)
Date: 12-06-06 06:33
When you just have one group and one behaviuoral measure, as I understand, there is no
SVD. The brain latent variables are simply the same as the cross-correlation value up to a
scalar.
In the data, we just have one behavioural measure! but we can get supplementary information
on the subjects.
What is the best choice and how to justify it ?
-- get some more information about the subjects in the group and separate it in different categories (because in fact you know that among the subjects in the group there are patients and controls), to
be able to use effectively the SVD.
-- leave it like this and get one singular value with the brain latent variables equal to the croo-correlation
up to a scaling factor.
Thanks for your answer,
Best wishes,
Sophie.
Date: 12-06-06 06:33
When you just have one group and one behaviuoral measure, as I understand, there is no
SVD. The brain latent variables are simply the same as the cross-correlation value up to a
scalar.
In the data, we just have one behavioural measure! but we can get supplementary information
on the subjects.
What is the best choice and how to justify it ?
-- get some more information about the subjects in the group and separate it in different categories (because in fact you know that among the subjects in the group there are patients and controls), to
be able to use effectively the SVD.
-- leave it like this and get one singular value with the brain latent variables equal to the croo-correlation
up to a scaling factor.
Thanks for your answer,
Best wishes,
Sophie.
Replies:
Re: One group One behavioural measure
Posted on 02/02/07 12:17:48
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 12-06-06 10:43
> When you just have one group and one behaviuoral measure, as I
> understand, there is no
> SVD. The brain latent variables are simply the same as the
> cross-correlation value up to a
> scalar.
In fact, we do perform SVD on one group and one behaviuoral measure.
> In the data, we just have one behavioural measure! but we can
> get supplementary information
> on the subjects.
>
> What is the best choice and how to justify it ?
>
> -- get some more information about the subjects in the group
> and separate it in different categories (because in fact you
> know that among the subjects in the group there are patients
> and controls), to
> be able to use effectively the SVD.
If you would like to see group effects, this would be the best choice.
> -- leave it like this and get one singular value with the brain
> latent variables equal to the croo-correlation
> up to a scaling factor.
Certainly, you can still leave it as one group, and perform SVD on one group and one behaviuoral measure.
Date: 12-06-06 10:43
> When you just have one group and one behaviuoral measure, as I
> understand, there is no
> SVD. The brain latent variables are simply the same as the
> cross-correlation value up to a
> scalar.
In fact, we do perform SVD on one group and one behaviuoral measure.
> In the data, we just have one behavioural measure! but we can
> get supplementary information
> on the subjects.
>
> What is the best choice and how to justify it ?
>
> -- get some more information about the subjects in the group
> and separate it in different categories (because in fact you
> know that among the subjects in the group there are patients
> and controls), to
> be able to use effectively the SVD.
If you would like to see group effects, this would be the best choice.
> -- leave it like this and get one singular value with the brain
> latent variables equal to the croo-correlation
> up to a scaling factor.
Certainly, you can still leave it as one group, and perform SVD on one group and one behaviuoral measure.
Re: One group One behavioural measure
Posted on 02/02/07 12:18:20
Number of posts: 100
Re: One group One behavioural measure
Author: Randy (---.rotman-baycrest.on.ca)
Date: 12-06-06 14:52
sophie wrote:
>
> What is the best choice and how to justify it ?
>
> -- get some more information about the subjects in the group
> and separate it in different categories (because in fact you
> know that among the subjects in the group there are patients
> and controls), to
> be able to use effectively the SVD.
>
> -- leave it like this and get one singular value with the brain
> latent variables equal to the croo-correlation
> up to a scaling factor.
>
The answer depends somewhat on how different the groups are on your behavior measure. If all patients have a zero score, for instance, and controls and nonzero, then it really doesn't help more than having a dichotomous group code. If it is more broadly distributed, then you could run it thru PLS and then take the brain scores from the analysis to see if the scores are different between groups. There is a slight bias here if the behavior itself differs between groups, but so long as there is overlap of the behavioral measure between groups you may be okay.
Date: 12-06-06 14:52
sophie wrote:
>
> What is the best choice and how to justify it ?
>
> -- get some more information about the subjects in the group
> and separate it in different categories (because in fact you
> know that among the subjects in the group there are patients
> and controls), to
> be able to use effectively the SVD.
>
> -- leave it like this and get one singular value with the brain
> latent variables equal to the croo-correlation
> up to a scaling factor.
>
The answer depends somewhat on how different the groups are on your behavior measure. If all patients have a zero score, for instance, and controls and nonzero, then it really doesn't help more than having a dichotomous group code. If it is more broadly distributed, then you could run it thru PLS and then take the brain scores from the analysis to see if the scores are different between groups. There is a slight bias here if the behavior itself differs between groups, but so long as there is overlap of the behavioral measure between groups you may be okay.
Login to reply to this topic.
