Dear users of metaSEM,

I was wondering if someone could help me out with calculating partial correlations in stage 1 of TSSEM in the metaSEM package.

Here is the description:

Let’s say we have 3 variables (A, B, C) in our path model. A is the IV, B is the mediator, and C is the DV. When I run TSSEM stage 1, it will result in pooling a matrix of pairwise correlations. So far so good, no problem here.

Now, I introduce variable D because I want to control for omitted variable bias. I know that I can run stage 2 of TSSEM by specifying a direct effect of D on A and of D on C to control for it. So far so good, no problem here neither.

Additionally, I want to check whether the partial correlation of B-C after introducing control variable D significantly differs from the pairwise correlation of B-C in stage 1 of the analysis. So far not so good, here is the problem. Is there any method/code to calculate the matrix of partial correlations in TSSEM?

Best,

Thomas

Dear Thomas,

The partial correlation of B_C after controlling D is essentially the correlation between the residuals of B and C after controlling D. You may refer to Figure 3 in Preacher (2006).

Preacher, K. J. (2006). Testing complex correlational hypotheses with structural equation models. Structural Equation Modeling: A Multidisciplinary Journal, 13(4), 520–543. https://doi.org/10.1207/s15328007sem1304_2

Best,

Mike

Hello,

Similarly like TDJ I have a mediation model and I wan't to run meta analysis on the indirect effect. I can meta-analyze the indirect effect without covariates (following the example in Cheung, M, W.-L. (2015). Meta-Analysis: A Structural Equation Modeling Approach. ) However, in the original studies the mediation model has two covariates. How can I include the covariates in the analysis? Should I pool the partial correlation matrices for the independent variable, the mediator and the dependent variable, controlling for the covariates? or I pool the bivariate correlations of the five variables (in this case I am not sure I specify the S3 and A3 matrix right). Could you direct me to some resources where I can find the answer to my question? Thanks in advance

Hi Ana,

You may meta-analyze the five-variable correlation matrices and fit a model with the covariates correlated with the variables. Attached is an example.

Mike

Hi Mike, Thank you for the example! I think I managed to do what I want to do (meta-analyze the indirect effect, while taking into account the covariates). Would you mind having a look at the code to see if I indeed got it right. Thanks

It looks correct to me.