Print

Print
Hi Silvia -

On 28 April 2011 11:40, Silvia Juanes <[log in to unmask]> wrote:

Dear all,


I have a question regarding Glm contrasts,

We propose the following analysis:

We have two groups and a continuous variable and we want to know whether there is a greater correlation in a group compared to the other.


My EV's are:

EV1--> the control group,
EV2--> the patient group,
EV3--> the variable for controls,
EV4--> the variable for patients.


The design matrix is:

1 0 x 0
1 0 x 0
. . . 

0 1 0 x
0 1 0 x
. . .


The contrasts that we want to test are:

0  0 1 -1   - (because we want to see if the correlation is more significative in controls vs patient)


1 -1 1 -1   - (I think this contrast means if the correlation is more positive in controls compared to patients)


No, this second contrast doesn't seem right.  The first contrast 0 0 1 -1 tells you where the regression coefficient is larger in controls, and flipping the sign -  0 0 -1 1  - will tell you where the reverse is true.  

Eugene
 

However, in the last contrast, in addition to the effect that we are testing, there is another effect that is "confounded": whether the correlation is more negative in the patient's group compared to the control's group. This effect is not a problem for my hypothesis but what i would like to know is whether the second contrast is correct in order to test what we have explained.

 
Many thanks to you all,

Silvia