Binary variable and indirect effects ... PreviousNext
Mplus Discussion > Structural Equation Modeling >
 SeungYong Han posted on Saturday, July 22, 2017 - 6:56 am
Dear Dr. Linda Muthen & Dr. Bengt Muthen,

I have a path model exploring the associations among social network (two variables: X1 and X2), social support (X3), physical activity (X4), and depression (Y).

I am mainly interested in estimating the indirect effects of social network on depression through social support and physical activity:
- path 1: X1->X3->X4->Y
- path 2: X2->X3->X4->Y

Y on X3 X4
X4 on X3
X3 on X1 X2

The problem is, X4 is binary (1 or 0).
I read some questions and answers on this discussion board, but I am still not sure what to do.

I want to clarify the code to estimate indirect effects as well as direct effects. I would really appreciate your comments on below questions.

Q1) Is X4 (physical activity) a predictor or an outcome? My understanding is that it is both. Then, I wonder if I need to use the command “categorical are X4” under Variable?

Q2) Do I need to use WLSMV as the estimator to estimate indirect effects and use ML to compare the fitness across models?

Q3) Do I need to use Model Constraint instead of Model Indirect if I use WLSMV?

Q4) Last, what should I do if X4 is ordinal (e.g., 1=less than 1 day, 2=between 1 and 3 days, 3=between 4 and 6 days, 4=everyday)?
 Bengt O. Muthen posted on Sunday, July 23, 2017 - 5:27 pm
The simplest approach for you is to

Say Categorical = x4

Use Estimator = WLSMV

Use Model Indirect
 SeungYong Han posted on Monday, July 24, 2017 - 6:52 am
Thank you so much, Dr. Muthen.

Two more questions based on your answers.

1) Do I still use "categorical=x4" when x4 is ordinal? (1-2-3-4)?

2) Do I use ML if I want to compare the model fitness of the same model, for example, between male and female?
 Bengt O. Muthen posted on Monday, July 24, 2017 - 2:53 pm

No, stay with WLSMV.
 SeungYong Han posted on Wednesday, July 26, 2017 - 8:24 am
Thank you, again!
One more Q.

I want to add covariates in my model.
I have one continuous (age) and one categorical (obese or not).

I wonder what's the difference between
1) controlling for age and obesity in each equation
2) allowing correlations between them and the other variables excluding the dependent variable.

(Case 1)
Y on X3 X4 age obese
X4 on X3 age obese
X3 on X1 X2 age obese

(Case 2)
Y on X3 X4
X4 on X3
X3 on X1 X2
X1 X2 X3 X4 WITH age obese

I think we do 1) if age and obese actually have impacts on Y, X4 and X3 and do 2) if I am not sure about the relationships between them and X1-X4.

Last, In 2), I wonder if it is ok to get the correlation between obese (1/0) and continuous outcome variables? I get a warning message.

I would appreciate any advice!
 Bengt O. Muthen posted on Wednesday, July 26, 2017 - 3:49 pm
Approach 1 is the way to go. You have 4 covariates (variables not on the left-hand side of ON): x1, x2, age, obese.

Covariates should not be on a Categorical list - this list is only for DVs.
 SeungYong Han posted on Friday, July 28, 2017 - 6:29 am
Thank you so much, Dr. Muthen :-)
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message