I have never done a mediation analysis before, and I have also never used Mplus before, so I could use some guidance!
My code is currently structured as follows:
VARIABLE: NAMES are (I list many variables here)... USEVARIABLES are x1 x2 x3 y m; CATEGORICAL are y m; ANALYSIS: bootstrap=5000; MODEL: m ON x1 x2 x3; y ON x1 x2 x3 m; MODEL INDIRECT: y IND m; OUTPUT: sampstat interval(bootstrap);
x is a 4 level categorical variable dummy coded to 3 levels, m is binary, and y is a categorical variable with 3 levels.
The code runs fine with no warnings, but I am concerned because for the indirect effects, the estimates are: Estimate SE Est./SE p-value 0.000 0.000 999.000 0.000
Any advice you can offer would be greatly appreciated!
Take a look at the Model Indirect topic described in the V8 UG on or website. A common form is to say Y IND m x; Your input is lacking the x. To understand the mediation topic well in the context of categorical y and m, read our book Regression and Mediation Analysis using Mplus, especially chapter 8 which has several examples (Mplus scripts are on our website).
Thank you for your help! Our team has realized that our outcome needs to be treated as nominal, but if we use the NOMINAL= option, we cannot get indirect effects. Can you tell me if there is a way to code this?
Also, so far we are not seeing a very large mediation effect and we are wondering if there are any guidelines as to how large an effect size should be in order to determine if a mediation analysis is the right way to go on this project.
So you are saying that y is not ordinal but nominal? That complicates matters. One way is to dichotomize y in the different ways possible and run one mediation analysis at a time. If you have not done mediation analysis before you want to read up on it.
Regarding your last question, try asking on SEMNET.