Hi all.
I am doing a Multiple Linear Regression using SPM12 in Matlab 2020b and I have a question about defining contrasts (2nd level analysis>Factorial design specification> Estimation> Contrasts step).
I want to test positive and negative effects (because I expect both positive and negative regression slopes) for e.g., 2 factors (age and gene). I am inserting these variables in the covariates menu and I want to specify certain covariates of interest using contrast weights of 1 and covariates of no interest using contrast weights of 0.
When I inspect the design matrix, the first parameter is an intercept, the second parameter is age and the third parameter is gene. So I set up my contrasts in this way:
Age negative effect: 0 -1 0 (modelling only age, but its negative effect on the dependent variable of interest). Maybe modelling is not the best term, but specifying it as a covariate of interest.
Age positive effect: 0 1 0 (modelling only age, but its positive effect on the dependent variable of interest)
Gene negative effect: 0 0 -1 (modelling only gene, but its negative effect on the dependent variable of interest)
Gene positive effect: 0 0 1 (modelling only gene, but its positive effect on the dependent variable of interest)
I wonder if/how I could define the contrasts in a way that compares the two independent variables age vs. gene to one another. For example:
Age>Gene to be defined as 0 1 -1 (0 for the intercept; 1 for age because I want to see its effect over gene; -1 for gene to treat it as covariate of no interest).
Or indeed have the opposite scenario of Gene>Age (0 -1 1). However, I am not sure the output of these contrasts makes sense and how I can interpret it. Is this a correct way of going about this contrast?
Lastly, I wonder if I can test the combined negative or positive linear effect of both variables age and gene by specifying the following contrasts:
0 1 1 for combined positive linear effect
0 -1 -1 for combined negative linear effect
This doesn't seem correct to me or perhaps it is not optimal to do it in the context of MLR?
Any thoughts and advice on this will be very much appreciated.
With kind regards!
|