Confounder adjustment in mediation analysis

[Lauren Blockmans]

Dear lavaan users, 

I'm looking for advice on our mediation model with:

- predictor X (continuous variable), the volume of a specific brain region (Heschl's gyrus)

- mediator M (continuous variable), a behavioral measure of risk for reading difficulties

- outcome Y (continuous variable), the behaviorally measured reading ability. 

Since the value of predictor X is known to be very dependent on total brain volume, we would like to control for this (by including total brain volume as a covariate C). However, theoretically it does not make a lot of sense to also include the total brain volume as a covariate on M or Y. Could I use the model syntax below (only including the covariate C on X), or would that not have the desired effect (i.e. controlling for total brain volume when investigating the volume of one specific brain region)?

Y~ c*X
M ~ a*X
Y~ b*M

X ~ C

ab := a*b
total := c + (a*b)

Kind regards,

Lauren 

[Terrence Jorgensen]

You only need to regress endogenous variables (M and Y) on the confounder, which can be correlated with X.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen

[Wen L.]

Is C assumed to be correlated with only X, and neither M nor Y? If so, your direct and indirect effects will remain the same irrespective of whether you include 'X~C' in the model.

Wen Wei Loh
Assistant Professor | Department of Quantitative Theory and Methods, Emory University