Indirect effects in two-level logisti... PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Lesa Hoffman posted on Tuesday, September 30, 2008 - 2:36 pm
Greetings,

I have a proposed two-level model for a dichotomous outcome. I do not expect random slopes, only a random intercept. I am interested in multiple mediational effects (e.g., tests of indirect effects such as X --> M --> Y) at the between level. In reading through the version 5 manual, it appears I cannot use the MODEL INDIRECT command to test indirect effects when numeric integration is required. Further, bootstrapped SE’s and CI’s (that are often recommended when examining indirect effects) are not available with TYPE=TWOLEVEL.

Given that I am trying to use a two-level model for a dichotomous outcome, is there any kind of estimator or alternative procedure I can use in Mplus v. 5.1 in order to obtain formal tests of indirect effects (and contrasts between indirect effects) at the between level? I suppose one could model the level-2 aggregates of level-1 variables in a single-level analysis (although the outcome would still be dichotomous), but I wanted to see if there was a more principled way of doing this within the two-level procedure.

Thank you,
Lesa Hoffman
 Linda K. Muthen posted on Wednesday, October 01, 2008 - 8:42 am
You would need to use MODEL CONSTRAINT to define that parameter and obtain a standard error for it.
 Lesa Hoffman posted on Thursday, October 02, 2008 - 10:20 am
Hi Linda,

Thank you for your prompt and helpful response. I have tried to use the MODEL CONSTRAINT command to define an indirect effect as suggested (relevant part of syntax):

%BETWEEN%
Y ON X*;
M ON X* (XM);
Y ON M* (MY);

MODEL CONSTRAINT:
NEW (med*0);
med=XM*MY;

The estimate for the 'med' indirect effect in the output is not the value from multiplying the estimated XM (=.357)and MY (=.459) effects together (=.164), but something much smaller (=-.012). Can you tell what I am doing wrong?

Thank you again,
Lesa Hoffman
 Linda K. Muthen posted on Thursday, October 02, 2008 - 10:31 am
Please send the input, data, output, and license number to support@statmodel.com.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: