There are two ways of doing Monte Carlo simulation studies using Mplus.
One way is 1) to generate data externally using other softwares such as
SAS PROC IML and then analyze these data using Mplus (for details, visit 
http://www.statmodel.com/runutil.html).  Another way is 2) to generate
data internally using Mplus and also use Mplus to analyze these data. The 
simulation studies shown in this paper were done using the method 1),
but can also be done using the method 2).


1) Mplus input code for a single analysis using a single data set
(real data, or data generated outside Mplus)

TITLE: Tables 7 & 8 external
       CACE estimation with a single continuous outcome with one covariate.

       An example (Little & Yau, 1998) of CACE estimation for a single
       continuous outcome with detailed explanations can be found at 
       http://www.statmodel.com/mplus/examples/mixture/mix12.html.

       Other examples of CACE estimation for repeated measures or a categorical
       outcome can be found in Jo & Muthén (2001). Modeling of intervention
       effects with noncompliance: A latent variable modeling approach for
       randomized trials.  In G. A. Marcoulides and R. E. Schumacker (Eds.),
       Advanced Structural Equation Modeling: New Developments and Techniques
       (pp 57-87).  Lawrence Erlbaum Associates.

DATA:  file is one.data;
VARIABLE: names are outcome trtment covariat c1 c2;
          usev  are outcome trtment covariat c1 c2;
          classes = c(2);
          training = c1-c2;
ANALYSIS: type = mixture;
MODEL:
 %OVERALL%
     C#1 ON covariat;
     outcome ON trtment covariat;
     outcome;
    [outcome];

! noncomplier class
 %C#1%
     outcome;
    [outcome];
     outcome ON trtment@0 covariat;

! complier class
 %C#2%
     outcome;
    [outcome];
     outcome ON trtment covariat;


2) Mplus input code for a Mplus internal simulation

TITLE: Tables 7 & 8 internal
       CACE estimation with a single continuous outcome with one covariate.
       See Mplus manual for details of internal monte carlo simulations.
MONTECARLO: 
          names are u1 outcome trtment covariat;
          nobs = 300;
          nrep = 1000;
          seed = 1234567;
          nclasses = 2;
          gclasses = 2;
          categorical = u1(1);
          cutpoints = trtment(0.);
          missing = u1;
ANALYSIS: type = mixture missing;
MODEL:

        %OVERALL%
        [outcome*1.5];
        outcome ON trtment covariat;
        outcome*1;

        c#1 ON covariat*0;
        [c#1*0];

	%c#1%
        [u1$1@15];
        [outcome*1.5];
        outcome ON trtment@0 covariat*0.0;
        outcome*1.0;

        %c#2%
        [u1$1@-15];
        outcome ON trtment*-0.5 covariat*0.0;
        outcome*1.0;

MODEL MONTECARLO:

        %OVERALL%
        [trtment@0];
        trtment@1;
        [covariat@0];
        covariat@1;

        outcome ON trtment*0 ;

        c#1 ON covariat*0;
        [c#1*0];

	%c#1%
        [u1$1@15];
        [outcome*1.5];
        outcome ON trtment@0 covariat*0.0;
        outcome*1.0;

        %c#2%
        [u1$1@-15];
        [outcome*1.5];
        outcome ON trtment*-0.5 covariat*0.0;
        outcome*1.0;

MODEL MISSING:
        %OVERALL%
        [u1@15];
        u1 ON  trtment@-30;