Power analysis for GGMM with distal o... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Silvia S Martins  posted on Tuesday, October 10, 2006 - 12:24 pm
I'm a new M-Plus user.
I need to estimate power and sample size for a GGMM with a distal outcome. I've already run a montecarlo simulation using the values from my original dataset for the intercept and slope variances,the actual means for the different classes, the threshold starting point and the class prevalence expressed as a logit. What I want to get is the power analysis for a logistic regression of a binary outcome variable on a categorical predictor (i.e., class membership). What do I need to do after the montecarlo simulation to actually get the power and sample size for different Odds Ratios?
 Bengt O. Muthen posted on Tuesday, October 10, 2006 - 5:52 pm
Ex 8.6 has a distal categorical outcome u. On your Mplus CD, the corresponding Monte Carlo setup is included. You see there that the logistic regression relating u to the latent class variable c is handled by the threshold for u varying across the c classes. So with 2 c classes, that is the same as logistic regression with a 0/1 dummy predictor, shifting the logit intercept for u (the logit intercept is the negative of the threshold). It sounds like you want to know the power of rejecting that the 2 thresholds (the log odds) are the same. That can be done using Model Constraint and defining a "New" parameter:

New(t1t2);
t1t2= t1-t2;

where t1 and t2 are parameter labels for the two thresholds. Then the last column of the Monte Carlo output gives you the percentage rejection of t1t2 being zero, which is the estimated power.
 yuanjia wang posted on Tuesday, June 19, 2007 - 6:07 am
Dear Dr. Muthen:

I am trying to compute power for a GMM with count outcome. The outcome takes value 0, 1, 2,...6. I have prior data to estimate intercept and slope within each latent class. In the Monte Carlo simulations, I see the saved replication 1 outcomes take value 0, 1, 2,..., up to 100. How do I restrict the outcome to be less than or equal to 6? Do I need to provide cutpoints for the simulation? Thanks.
 Linda K. Muthen posted on Tuesday, June 19, 2007 - 8:21 am
It sounds like you are using some but not all of the parameters from your real data as population values in your Monte Carlo study. If you used them all, then I think you would obtain the range of the variable that you want. The CUTPOINTS option is only for covariates. See Example 11.7 where real data are used to obtain population values for a Monte Carlo study.
 yuanjia wang posted on Tuesday, June 19, 2007 - 8:44 am
Thanks for your reply. I did follow Example 11.7 and saved the output using "Estimates=datafile". I see in the 'Savedata' option there are "TYPE", "FORMAT" options. Is there anything else I should save?
 Linda K. Muthen posted on Tuesday, June 19, 2007 - 9:19 am
Did you then use datafile with the POPULATION and COVERAGE options in the Monte Carlo analysis? That is Step 2 of Example 11.7.
 yuanjia wang posted on Tuesday, June 19, 2007 - 9:25 am
Yes, I used both POPULATION and COVERAGE options, and referred them to the same file saved in the step 1.
 Linda K. Muthen posted on Tuesday, June 19, 2007 - 10:01 am
Then I would need you to send the inputs, data, outputs, and your license number to support@statmodel.com.
 Jon Elhai posted on Monday, March 31, 2008 - 12:46 pm
Drs. Muthen,
I have a basic growth model as in the User's Guide's example 6.1. However, I also have one predictor variable (x1) that is a binary "yes/no" variable thought to predict the intercept and slope. When trying to run a monte carlo study, I add the following syntax to, presumably, indicate that x1 is a binary variable with a cutpoint of 0: cutpoints = x1(0);

However, I get the following error that I'm hoping you can explain:
*** ERROR in Montecarlo command
Only x-variables may have cutpoints in this type of analysis.
Cutpoints for variable X1 were specified.
 Linda K. Muthen posted on Monday, March 31, 2008 - 2:31 pm
It sounds like you are mentioning the mean and variance of the variable x1 in the MODEL command. They should be mentioned only in the MODEL POPULATION command.
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: