Binary variable as a mediator and indirect effects estimation in SEM model by Mplus

I would really appreciate anyone's advice on my modeling in MPlus :)

So we are interested in estimating the indirect effects of X1 (t_health) and X2 (t_freq) on Y (cesd) through M (pa). PA is a binary bariable (1/0).

Q1) Should I treat PA as a categorical in "variable" command? This is a predictor as well as an outcome so I am a bit confused.

Q2) It seems like I need to use "WLSMV" not "ML" when I have a binary predictor in the model. Is this correct?

Q3) I am not sure if I need to use "MODEL INDIRECT" or "MODEL CONSTRAINT". Which one is the correct one?

Here is the Mplus code.

Variable:
...
categorical are pa;

Analysis:
type=general;
estimator=WLSMV;
iterations=5000;

Model:
cesd ON pa ssp;
pa ON ssp;
ssp ON t_health t_freq;

Model indirect:
cesd IND SSP;
cesd IND t_health;
cesd IND t_freq;

• This question seems off-topic here, because it asks about software-specific elements. If you rephrase it to ask about the statistical issues rather than Mplus application, it might be on topic. But I also recommend asking this question on Mplus Discussion, since Muthéns are quite responsive. – T.E.G. Jul 18 '17 at 17:29
• Thank you for the clarification. Yes, I will try Mplus discussion website though "Mplus Discussion is unavailable until July 21." I will wait! :) – Brainupgraded Jul 19 '17 at 15:54