Hi,

No, you should not mean center within group, that removes the ability of that regressor to adjust for mean differences due to age. 

See this for more details

http://mumford.fmripower.org/mean_centering/

Cheers,

Jeanette

On Fri, Dec 27, 2013 at 1:31 AM, Himanshu Joshi <[log in to unmask]> wrote:

Hello Jennate,

Season's greetings,

I want to generate design.mat and design.con file for dual regression and I have a confusion

If suppose I have two groups
first group mean age is X

and second group mean age is Y
for i=0......n
demeaning for first group is xi-X and demeaning for second group yi-Y
Shall I enter these demeaned values like this
or
shall I consider all the subject irrespective of which group they belong and demean accordingly and enter these demeaned values to their corresponding subjects in Glm.

On Fri, Mar 15, 2013 at 5:27 PM, Jeanette Mumford <[log in to unmask]> wrote:

The second design will create 4 contrasts (4 corrected p files), because you specified 4 contrasts.  Randomise doesn't produce anything by EV.

Jeanette

On Fri, Mar 15, 2013 at 6:48 AM, Himanshu Joshi <[log in to unmask]> wrote:

Thanks and I got it .

Let us discuss design 1.png, I guess it creates equal number of tstat files as it has no of explainatory vaiables not equal to number of contrast--Since it creates 2 tstat files tstat1 and tstat2 as it has two EV's. so what is the role of four contrast provided in the design?

please comment 

On Thu, Mar 14, 2013 at 8:34 PM, Jeanette Mumford <[log in to unmask]> wrote:

To understand the interpretation of a contrast, you only need to know what your parameters are and then multiply your contrast vector by the parameter vector.  In this case your parameters (estimates corresponding to the design matrix columns) are b1 and b2, where b1 is the mean for group 1 and b2 is the mean for your second group.  The hypothesis you are testing with a contrast c is a 1-sided hypothesis,

H0:  cB<=0   (null)

vs

H1:  cB>0     (alternative)

Where c is  your row-vector contrast.

First case your c=[1 -1], so your null hypothesis is

H0: [1 -1] * [b1 b2]'<=0   which multiples out to     b1-b2<=0

OR  (after adding b2 to both sides)

H0:   b1<=b2

Rejecting H0 means you are accepting the alternative, which is

H1:  b1 > b2

So, yes, rejecting the null for your first contrast [1 -1] means that b1 > b2.

Jeanette

On Thu, Mar 14, 2013 at 9:08 AM, Himanshu Joshi <[log in to unmask]> wrote:

Let us discuss the design matrix further 

Does design.png (see attached) gives results for activation of group1 greater than group2 ?

Is design1.png (see attached) is efficient enough for what it meant to say (see contrast)?

---------- Forwarded message ----------
From: Himanshu Joshi <[log in to unmask]>
Date: Tue, Mar 5, 2013 at 6:49 PM
Subject: Re: [FSL] Design matrix for dual_regression
To: Jeanette Mumford <[log in to unmask]>



Thanks Jennate,
Which design is correct, to avoid further confusion I attach both designs
and the mail from mark also I had forwarded before just before this mail
design or design1  ?

let us discuss both designs
I suppose in design 1 tstat 1 to tstat4 files will form but using matrix design 1produced only two tstat files.

Thanking you for your support

 

On Tue, Mar 5, 2013 at 6:22 PM, Jeanette Mumford <[log in to unmask]> wrote:

I couldn't find that response from Mark, but he would know better than I.  You still need to run at least 2 models to get at everything that you need as the 1/2 Group membership will not work for the group comparisons.

Cheers,

Jeanette

On Tue, Mar 5, 2013 at 1:31 AM, Himanshu Joshi <[log in to unmask]> wrote:

Hello ,

According to Mark's Mail to carole on 25th feb2013 the first two contrasts find the group differences in the initial design.png (attached on 4th march) with the numbers (1 and 2) assigned to the group 1 and group2.

Which one is correct earlier design.png (attached on 4th march) or newer design.png (attached on 5th march)

Thanking you in anticipation

On Tue, Mar 5, 2013 at 10:34 AM, Himanshu Joshi <[log in to unmask]> wrote:

Thanks Jenate,
I am attaching the rectified matrix as described by you
I will restrict groups to be swapped.
Does it will tell activations of control>patients (tstat1) and patients>control (tstat2) when I run using this design in ualregression

On Mon, Mar 4, 2013 at 6:27 PM, Jeanette Mumford <[log in to unmask]> wrote:

Hi,

For you group means I cannot be sure that randomise does the permutations properly when you give it a model like this as it would need to know to permute the data using random multiplication by 1's and -1's (somebody else can verify).  Instead I would run randomise on each group alone following the instructions for a 1-sample t-test to get your two group means.

For the group comparisons use your first design, but it needs some work.  Firstly, change the "group" column to all 1's.  In randomise this defines how the data will be swapped and you're basically not allowing the group labels to swap, when that's exactly what you need for the 2-sample t-test. Then use the [1 -1] and [-1 1] contrasts.

Cheers,

Jeanette

On Mon, Mar 4, 2013 at 12:53 AM, Himanshu Joshi <[log in to unmask]> wrote:

Yes Jeannette,
 You are correct no of rows must be more than number of columns in a design matrix.
Can I ask you one more doubt
If I have a control X patient data of 10 X 10 inputing the design matrix to dual regression with contrast value=2 and putting EV values 1 in a diagonal manner see design.png
Does tstat1 shows higher activations of control than patient data?
Does tstat2 shows higher activations of patient than control data?

I had designed it using contrast value=2 and f test=2 (see design1.png)
Does tstat1 shows group mean activations of control?
Does tstat2 shows group mean activations for patients?

Thanking you in advance

On Fri, Mar 1, 2013 at 7:04 PM, Jeanette Mumford <[log in to unmask]> wrote:

A design matrix must have more rows than columns for the model to be estimable, so you need to cut down on your DV.

Cheers,

Jeanette

On Fri, Mar 1, 2013 at 4:40 AM, Matteo Diano <[log in to unmask]> wrote:

Thank you for your answer.
I've tried to construct a matrix 3 subjects X 4 EV and I have got the same error... I am using FSL 5.0.2 on WMplayer linux centos 6 64bit.
I've tried also on MAC 10.7.5 and I've got the same error as well.
Any suggestions will be very appreciated!!!
Thanks you.
Regards

Matteo

2013/2/26 Stephen Smith <[log in to unmask]>

Hi - most likely this is an "illegal" design matrix, one way or another - maybe you have empty EVs or it's rank deficient in some other way.

Cheers.

On 20 Feb 2013, at 15:55, Matteo Diano wrote:

Dear FSL experts,
I have constructed a matrix with Glm that is composed by 24 subject X 30 EV (behavioral variables). When I tried to save it I received an error message:

Problem with processing the model: terminate called after throwing an instance of 'NEWMAT::ProgramException'


Glm saved me only a *.fsf but I need the *.mat *.con to use in the dual_regression analysis (randomise step). I noticed if I divide the matrix in two (15 and 15) I can save both of them...
Why I cannot save the entire matrix?  It is right if I use the divided matrix instead of the unique one for the dual_regression analysis?
Thank you very much for your helpfulness!
Regards

Matteo

---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------

--
Himanshu Joshi
M.Tech. Cognitive & Neuroscience.
Senior Research Fellow
Department of Psychiatry
NIMHANS, Bangalore

http://mbial.weebly.com/himanshu-joshi.html

--
Himanshu Joshi
M.Tech. Cognitive & Neuroscience.
Senior Research Fellow
Department of Psychiatry
NIMHANS, Bangalore

http://mbial.weebly.com/himanshu-joshi.html

--
Himanshu Joshi
M.Tech. Cognitive & Neuroscience.
Senior Research Fellow
Department of Psychiatry
NIMHANS, Bangalore

http://mbial.weebly.com/himanshu-joshi.html

--
Himanshu Joshi
M.Tech. Cognitive & Neuroscience.
Senior Research Fellow
Department of Psychiatry
NIMHANS, Bangalore

http://mbial.weebly.com/himanshu-joshi.html

--
Himanshu Joshi
M.Tech. Cognitive & Neuroscience.
Senior Research Fellow
Department of Psychiatry
NIMHANS, Bangalore

http://mbial.weebly.com/himanshu-joshi.html

--
Himanshu Joshi
M.Tech. Cognitive & Neuroscience.
Senior Research Fellow
Department of Psychiatry
NIMHANS, Bangalore

http://mbial.weebly.com/himanshu-joshi.html

--

Himanshu Joshi
M.Tech. Cognitive & Neuroscience.
Senior Research Fellow

Ph.D Scholar,

Department of Psychiatry
NIMHANS, Bangalore

http://mbial.weebly.com/himanshu-joshi.html