Path analysis, sample weights
Message/Author
 Binay Adhikari posted on Sunday, May 13, 2018 - 8:54 pm
Hello everyone,
I am new to mplus. I am working on a multiple mediator path model to examine mediation effect. All my endogenous variables are binary and I also have sample weights. I wanted to know how to interpret the coefficients of the regression results ( it seems to use probit link) and to use bootstrapping to estimate indirect effects.

Below is the code I used for path analysis:

DATA: FILE IS mplus_.csv;
VARIABLE: NAMES ARE wi pc ra wiq pcq gen inc ed age years
homeowner walk rwalk pa bmi dia
hbp hrtdis ma mh str weight abor asian sasian other;
CATEGORICAL ARE walk pa bmi dia;
MISSING ARE all(-99);
WEIGHT = weight;
USEVARIABLES dia wi ra pa walk bmi gen inc ed abor asian
sasian other years age;
ANALYSIS: PARAMETERIZATION = THETA;

MODEL: walk pa ON wi gen inc ed abor asian sasian other years age ra ;
pa ON wi walk gen inc ed abor asian sasian other years age ra ;
bmi ON wi pa walk gen inc ed abor asian sasian other years age ra ;
dia ON wi pa walk bmi gen inc ed abor asian sasian other years age ra ;
MODEL INDIRECT:
dia IND wi;
dia IND pa;
dia IND walk;
bmi ind wi;

Thanks!
 Bengt O. Muthen posted on Monday, May 14, 2018 - 4:45 pm
WLSMV gives effects for the continuous latent response variable behind the binary DV, so the effects are interpreted as for continuous outcomes. Note that this is different from effects for the binary observed variable itself - those effects are obtained via counterfactual definitions as described in our Topic 11 YouTube video on our website - and our RMA book. But they are more complex with multiple mediators as you see in the article:

Nguyen, T.Q., Webb-Vargas, Y., Koning, I.K. & Stuart, E.A. (2016). Causal mediation analysis with a binary outcome and multiple continuous or ordinal mediators: Simulations and application to an alcohol intervention. Structural Equation Modeling: A Multidisciplinary Journal, 23:3, 368-383 DOI: 10.1080/10705511.2015.1062730
 Binay Adhikari posted on Wednesday, May 16, 2018 - 11:21 am
Thanks Bengt for resources. They are really helpful

I have one more follow up question. It seem complicated to interpret the coefficient with binary variables as mediation. As a way around for interpretation, if I want to calculate the relative/percentage effect of each pathway, can I calculate the percentage directly from the total, direct, indirect, and specific indiret effect from the results of the path model?

Thanks again.
 Bengt O. Muthen posted on Wednesday, May 16, 2018 - 1:43 pm
That's a question for SEMNET.
 Binay Adhikari posted on Friday, May 18, 2018 - 11:16 am
Thanks Bengt,

Just wanted to confirm. As I see from the result, the total effect seems to be a simple sum of all the direct and indirect effects. I was wondering, if the total effect is sum of all the components then, can I use the total, direct, indirect, and specific indirect effects that I get from Mplus to calculate the percentage effect of each pathway. I understand this may be relevant for different forum but I also wanted to understand the "Model Indirect" command results.

Thanks
 Bengt O. Muthen posted on Friday, May 18, 2018 - 1:50 pm
You can do that.