Pairwise Contrasts in Multiple Mediat...
Message/Author
 Estee posted on Friday, October 04, 2013 - 8:57 am
Hi,

I am using mplus to run an analyses of a multiple mediator model (3 mediators). I created pairwise contrast such as below:

Con1 = a1b1 - a2b2
Con2 = a1b1 - a3b3
Con 3 = a2b2 - a3b3

In the output, the estimates of these contrasts appear at the part of "Model Results" which is before "Standardized Model Results". So, I am wondering the estimates for these contrasts in the output is standardized or unstandardized? Is it possible to get standardized estimates for these contrasts?
 Linda K. Muthen posted on Friday, October 04, 2013 - 10:37 am
If you created these in MODEL CONSTRAINT, they are unstandardized.
 Estee posted on Friday, October 04, 2013 - 10:56 am
Dear Dr. Linda,
so, how can I get standardized estimates for these contrasts?
 Linda K. Muthen posted on Friday, October 04, 2013 - 1:24 pm
You would need to do this using model constraint. You would need to specify the standard deviations for con1, con2, and con3 and specify the standardized coefficients.
 Estee posted on Friday, October 04, 2013 - 3:42 pm
How to specify standard deviations and standardized coefficients? where to get these values? Example of syntax?
TQ.
 Bengt O. Muthen posted on Sunday, October 06, 2013 - 9:56 am
First you have to decide if you want contrasts based on standardized coefficients or if you want a standardized contrast (divided by its SD). Let's assume the former (the latter is automatically obtained as the z score in Model Constraint). You are showing formulas with a*b which implies that you want to standardize a*b wrt the SDs of X and Y in the X->M->Y chain. So you need to be able to express the variance of Y from your model parameters. You do this by labeling parameters in the Model command, and then use those labels to computed the Y variance in Model Constraint. The X variance is obtained simply by the Model command labeling:

x (xvar);
 Jenna E Finch posted on Thursday, March 24, 2016 - 5:10 pm
Hello,

We are hoping to contrast several sets of indirect pathways in a SEM with one predictor, 4 mediators, and 3 outcomes. There are two mediators at time 1 and then the same mediators at time 2.

We have calculated the indirect effect of the X --> MedA at time 1 --> MedA at time 2 --> Y using the VIA command, but we are unsure how to compare this to the indirect effect of X --> MedB at time 1 --> MedB at time 2 --> Y, since there are two mediators.

Thank you so much for your help.
 Bengt O. Muthen posted on Thursday, March 24, 2016 - 6:28 pm
You can express the two indirect effects in Model Constraint and also their difference. It will then be tested for equality.
 Jenna E Finch posted on Thursday, March 24, 2016 - 10:54 pm
Bengt,

Thank you for your response. I apologize for not being clear above and I have a quick follow-up question.

For a model with one mediator (e.g. X --> M --> Y), you would specify the indirect effect as shown below:
MODEL:
y ON m (b1);
m ON x (a1);

MODEL CONSTRAINT:
NEW(a1b1);
a1b1=a1*b1

For a model with two mediators, how would you express this indirect effect? Is this correct for model X--> M --> L --> Y?

MODEL:
y ON l (c1);
l ON m (b1);
m ON x (a1);

MODEL CONSTRAINT:
NEW(a1b1c1);
a1b1c1=a1*b1*c1

Or are the indirect effects not calculated with the product of the betas of the pathways when more than one mediator is included?