I have a question regarding the linear-model specification. If I am meta-analysing two different psychological constructs (for example A and B), and I want to examine their contrasts (i.e., A>B and B>A), should I be including a dummy variable in the linear model where A=1 and B=0? In this case the results of the linear model would show A>B (for positive) and B>A for negative?
Is this understanding correct? Or do I have to do a separate linear model with a dummy variable where B=1 and A=0 for finding the results for B>A?

In my data, I see different results from the linear model depending on whether I specify A=1 or B=1.

I guess in this case I should report A>B as the results when A=1, and B=0; and B>A as the results obtained from B=1, A=0?


Appreciate your help, thank you very much.
Best,
Avi

Dear Avi,

If you want to perform the contrasts A > B and B > A for your linear model; you have to add both variables in the model and specify the contrast A=1 and B=-1 (A>B); and separately, the contrast A=-1 and B=1 (B>A).

If you set one of the variables to 0, the model will control for the effect of that variable but will not be included in the contrast (eg. A=1 and B=0, the contrast will be if A=0)

Kind regards,
Lydia

Hi,

I have tried doing what you suggested but I get this error: gsl: ../../gsl-2.5/linalg/lu.c:266: ERROR: matrix is singular Default GSL error handler invoked. gsl: ../../gsl-2.5/linalg/lu.c:266: ERROR: matrix is singular Default GSL error handler invoked.

Basically, I have two columns (different subsets of data). ON column A, I have values 1 for a certain subgroup and 0 for the other. Similarly, in column B I have 1 and 0 according tho whether the row is in that subset of studies.

I want to look at difference between the subgroups (A-B) and (B-A). As you suggested, I put 1 and -1, respectively, in the linear model specification for these two columns.

I get the above error.

I am attaching my SDM table, Please advice,
Avi