Hi metaSEM users, hi Mike,
I am just wondering if and how to apply a moderator analysis in metaSEM. Can I apply the approach you describe in your book beginning on page 170? I would like to test a moderating effect between two effect sizes by treating one variable as moderator. Or is there any other approach?
Thanks for your help.
Johannes
Hi Johannes,
Unfortunately, it has not been implemented in metaSEM yet. It can be done in OpenMx as shown in the examples https://courses.nus.edu.sg/course/psycwlm/internet/metaSEMbook/index.html#mediation-and-moderation-models
If you have problems in implementing it in OpenMx, please feel free to post them here.
Mike
Hi Mike,
Thanks for the link. I tried to follow the analysis and to specify a moderator analysis for my data set.
I have some questions on that:
- Can I use a correlation matrix as input?
- What about the covariance matrix and number of observations?
- How do i specify the M matrix correctly?
I attached my R file with what I got so far.
Thanks for your help
Johannes
Hi Johannes,
A single correlation matrix is not sufficient for modelling the moderating effect between effect sizes. We need to have the effect sizes (y) and their sampling covariance matrices (V) as input; otherwise, we cannot specify the mean structure (M matrix in the RAM specification).
By the way, I wonder whether you are testing something similar to the interaction between x and m on y in regression analysis, e.g., y~x+m+x:m or y~x*m. If this is the case, it is not possible to test the interaction with only a correlation matrix on x, m and y.
Mike
Hi Mike,
well I have the primary studies' correlation matrices I also use for the TSSEM/three-level analysis. Does it work with these? Or is it too complicated to calculate all covariances matrices first?
Johannes
Hi Mike,
I would like to ask again about the possibilities to perform a moderator analysis in a meta-analysis with MASEM/TSSEM because I am not sure from reading your book.
For example, my sample consists of 3 variables A, B and C and I gathered the correlations between all three variables from studies. Now I would like to examine the effect of A on C and how B is moderating that effect.
Is that possible? What data do I need? Or can I only use study characteristics as moderating variables?
I really appreciate your help on that.
Johannes
Hi Johannes,
As far as I know, it is not possible to test the moderation with only the correlation matrix among the variables.
Using study characteristics as moderators is well developed in meta-analysis. However, I am not aware of any good methods to address the moderation within the studies. This is an area that may need further attention.
Mike
Hi Mike,
Thanks for you reply. That means that you are always referring to study characteristics as moderators in your book?
Johannes
Hi Johannes,
Yes.
Mike