

Indirect effects covariance matrix 

Message/Author 


We are doing several crosslagged path analyses that include three waves of data. We also are using 10 multiplyimputed datasets. We've conducted our analyses, and now want to do some tests of mediation. Using Mplus's "INDIRECT" command, would usually be no problem and easy enough to set up, however, Mplus does not allow the user to use the "INDIRECT" command when doing an IMPUTATION style analysis. Our solution to this problem is to run the analysis on each imputed dataset individuallly, and then average the results across the 10 datasets. This is completely doable for the parameter estimates, but because the standard errors have to be averaged in a more complicated manner, we were hoping to do this using Stata's or Sas's matrix algebra program to complete the math that needs to be done. We know that using Mplus's "SAVEDATA" command and then using the "RESULTS ARE" subcommand will save the parameter estimates, standard errors, and fit statistics to a new .dat file, but what we really need are the covariances for all of the direct and indirect effects, and their related standard errors in a matrix format. Is there any way that Mplus outputs/creates a datafile of these covariances so that we can use them in SAS or STATA? 


I think an easier solution which I just thought of and tried is to use MODEL CONSTRAINT to define the indirect effects. Then you will obtain a standard error automatically. 


Thank you for your reply. I have reviewed the Mplus User's Guide to examine the MODEL CONSTRAINT command, but was wondering if you could give me a brief example of specifying an indirect effect in the MODEL CONSTRAINT command just to be sure that I understand what I have read. Thank you for your time. Andrea 


In the MODEL command, you need to give labels to the two regression coefficients involved in the indirect effect. In the MODEL CONSTRAINT command, you need to specify a new parameter and make that new parameter equal to the product of the two regression coefficients. 

Back to top 

