Power analysis for Latent Profile Ana... PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Sarah Vogel posted on Friday, May 29, 2020 - 10:50 am
I am trying to run a power analysis for an LPA. My sample size is 300. I imagine the best way to do this would be a Monte Carlo simulation, but I'm not sure of how to go about doing that. Is there any documentation available to help get this started?

Many thanks
 Bengt O. Muthen posted on Saturday, May 30, 2020 - 11:45 am
Choose a UG example close to the LPA you want and look at its corresponding Monte Carlo run which is also posted on our website when you go into the examples via

http://www.statmodel.com/ugexcerpts.shtml
 Yu Hui Zhang posted on Sunday, September 27, 2020 - 8:13 pm
I would like to ask a few follow-up questions.

1. I looked at

mcex7.9 and mcex7.10

and in terms of power analysis, would one look at 95% Cover and % Sig Coeff of the means, variances and Means C#1 to assess power? The % Sig Coeff for Means C#1 is 0.000, would it be okay?

2. Is there a way that one could specify item correlations? I added Y1 with Y4@.1 but the program does not seem to take it when I looked at the sample statistics.

Thank you!
 Yu Hui Zhang posted on Sunday, September 27, 2020 - 9:12 pm
3. I would also like to ask if there is a way to specify the degree of separation among profiles? Much thanks!
 Bengt O. Muthen posted on Monday, September 28, 2020 - 5:13 pm
1. You have to decide what you want to know the power of. I don't think the quantities you mention make sense - for instance, [c#1] is just the logit corresponding to the class proportion - do you want to know the power of it being different from 0.5 or zero? More relevant power would be class differences in e.g. the class-specific means which you can express in Model Constraint.

2. See UG ex 7.16.
 Yu Hui Zhang posted on Tuesday, September 29, 2020 - 8:09 am
Thanks, Dr. Muthen! Yes, I do, and thanks for pointing me to UG ex 7.16.

I ran UG ex 7.9 this time I requested tech 11 and tech 14. I would like to ask

1. Whether one could interpret the significant V-L-M-R Likelihood results (<0.5) averaged over replications as indicating that the study had enough power or capacity to correctly recover a two-class vs. a one-class solution?

2. The proportion of replications rejected at the 5% level is the same as power?

3. If one wants to compare a two- (the population) to a three-class solution, then one would have to use the data sets generated and run Mplus on each of them and tabulate and summarize the results as the Monte Carlo run will only compare the true (population) to the no. of class minus one solution?

Thanks a lot!
 Tihomir Asparouhov posted on Wednesday, September 30, 2020 - 8:17 am
1 & 2. Yes

3. Yes but that process is automated in Mplus. You can just change the input to this
genclasses = c(2);
classes = c(3);
 Yu Hui Zhang posted on Thursday, October 01, 2020 - 8:33 pm
Thank you very much, Dr. Asparouhov, for the replies and the very helpful tip!
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: