mediation analysis involving multiple mediators (serial and parallel)

[Marion Bailhache]

Hello, 

I reformulate correctly my question.

can we correctly estimate indirect (and direct) effects when mediators are involved in several causal paths using lavaan package ?

For example, in this example (figure), can we accurately estimate the indirect effect of exposure on behavior 1 via the different mediators (1, 2, 3 and 4), and the indirect effect of exposure on behavior 2 via the different mediators (1, 2, 3 and 4) ?

Thank you very much for your help and insights.

Marion

PS : Can you delete my previous post :"mediation analysis involving both parallel and serial...", which was confusing and erroneous? thank you very much

[Shu Fai Cheung (張樹輝)]

May I ask a few questions regarding the model?

- Is the exposure variable a categorical variable with more than two categories?

- Will all the medaitors be treated as continuous?

- Will the two outcome variables, Behavior 1 and Behavior 2, be treated as continuous? Or will they be treated as ordinal?

-- Shu Fai

[Marion Bailhache]

Hello,

_ the exposure variable is a categorial variable with five categories.

_ all the mediators are continuous.

_ In my model, I have in fact five outcome variables, with four outcomes variables, that are treated as continuous, and one outcome variable that is a categorial variable withe two categories.

I use the estimator "WLSMV" with parameterization theta.

The fit measures of the model are good : cfi 0,978, tli 0,98, rmsea 0,021, srmr 0,003

However, I have concerns about the accuracy of estimating the different indirect effects of exposure to different outcomes and about the validity of these tests, notably because mediators are involved in different causal path, and outcomes are not all treated as continuous variable.

Marion