MODEL INDIRECT or MODEL CONSTRAINT?
Message/Author
 Jinni Su posted on Friday, September 27, 2013 - 1:33 pm
Hello,

I want to examine a SEM model where X is hypothesized to influence Y through multiple indirect pathways that involve multiple mediators. Specifically, it is hypothesized that:

X-->M1-->Y
X-->M2-->Y
X-->M1-->M3-->Y
X-->M2-->M3-->Y
X-->M1-->M2--M3-->Y

How do I specify a model to examine all the specific indirect effects and total indirect effects of X on Y? Are the following commands correct?

MODEL INDIRECT:
Y IND M1 X;
Y IND M2 X;
Y IND M1 M3 X;
Y IND M2 M3 X;
Y IND M1 M2 M3 X;
Y IND X;

Or should I use MODEL CONSTRAINT instead?

MODEL CONSTRAINT:
NEW(XM1Y XM2Y XM1M3Y XM2MM3Y XM1M2M3Y);
XM1Y = XM1 * M1Y;
XM2Y = XM2 * M2Y;
XM1M3Y = XM1 * M1M3 * M3Y;
XM2M3Y = XM2 * M2M3 * M3Y;
XM1M2M3Y = XM1 * M1M2 * M2M3 * M3Y;

How do I obtain total indirect effects of X on Y if I use MODEL CONSTRAINT?

Thanks!
Jinni
 Linda K. Muthen posted on Saturday, September 28, 2013 - 6:39 am
If you say

MODEL INDRIECT:
y IND x;

you will obtain tests for all indirect effects from x to y.

If this is not available, I need to know more about the scales of your variables.
 Jinni Su posted on Sunday, September 29, 2013 - 7:50 am
Thank you very much! I think this will work.
 Jonathan Jong posted on Saturday, May 10, 2014 - 8:54 am
I'd like to be able to do this too! But I'm new to MPlus and am wondering what to type in. I have:

title: mediation;

data: file = data.dat;

variable:
names = X M1 M2 Y;
usevar = X M1 M2 Y;
missing = all(-999);

analysis:
type = general;
bootstrap = 5000;

output: cinterval (bcbootstrap);

model:
Y on M1 M2 X;
M1 on X;
M2 on X;
MI with M2;

model indirect:
Y ind X
 Linda K. Muthen posted on Saturday, May 10, 2014 - 11:21 am
This looks correct. Give it a try.
 Jonathan Jong posted on Saturday, May 10, 2014 - 2:49 pm
I don't seem to get X-M1-M2-Y in the output, just the two specific indirect effects separately. Am I looking in the wrong place in the output? I'm quite new to this.
 Linda K. Muthen posted on Saturday, May 10, 2014 - 5:22 pm
You need

m2 ON m1;

you have

m1 WITH m2;
 Jonathan Jong posted on Sunday, May 11, 2014 - 1:00 am
Ah, thanks Prof.!

I suppose if I want to add an M3 into the chain, then I would need:

model:
Y on M1 M2 M3 X;
M1 on X;
M2 on X;
M3 on X;
M3 on M2;
M2 on M1;

Is that right?
 Linda K. Muthen posted on Sunday, May 11, 2014 - 7:01 am
You use an ON statement for whichever regression you want.
 Fredrik Falkenström posted on Monday, July 08, 2019 - 8:35 am
I have a similar problem as above, but my model is a cross-lagged panel model with 8 time-points and I am trying to get the total effect of X at times 1-7 on Y at t=8 (the last one). The cross-lags and autoregression only have one lag, but there are hundreds of indirect effects - too many to specify easily using Model Constraint. Is there any way of doing this using the Model Indirect command? I would also, in the next step, want to calculate the difference in this total effect between two groups (I am using latent class analysis with known classes, using Estimator = Bayes).

Best,

Fredrik Falkenström
 Bengt O. Muthen posted on Sunday, July 14, 2019 - 11:43 am
 Fredrik Falkenström posted on Tuesday, July 16, 2019 - 11:22 am
Sorry for being unclear, but my problem is that I don't know if what I want to do can be done at all. I would like to add the total indirect effects of X1 -> Y8, X2 -> Y8, X3 -> Y8, X4 -> Y8, X5 -> Y8, and X6 -> Y8, plus the direct effect of X7 -> Y8, to get the total effect of X on Y8. As far as I know it is only possible to calculate these separately using the Model Indirect, i.e:

Model Indirect:
Y8 IND X1;
Y8 IND X2;
Y8 IND X3;
etc.

If I instead were to use Model Constraint, I would need to specify each and every indirect path which would take a lot of time and space, plus the risk of missing some of them.
 Bengt O. Muthen posted on Tuesday, July 16, 2019 - 6:12 pm
I don't think that sum of effects can be obtained via Model Indirect.
 Fredrik Falkenström posted on Thursday, July 18, 2019 - 11:03 am
Ok, I see. Is there a formula for getting the standard errors for the sum of two (or more) coefficients that I can use to calculate this afterwards?
 Bengt O. Muthen posted on Thursday, July 18, 2019 - 1:04 pm
You can use the same Delta method approach that Model Constraint uses. I like the description in Kendall-Stuart. The variances and covariances of parameter estimates are given in Tech3.
 Alexander O'Donnell posted on Thursday, May 28, 2020 - 5:22 pm
Hello, I am looking to examine moderated mediation using the XWITH command, with TYPE=RANDOM. Is the only way to obtain the indirect effects with the MODEL CONSTRAINT process?

Many thanks,
Alex.
 Bengt O. Muthen posted on Saturday, May 30, 2020 - 11:41 am
Yes.