Sample Size for LCA PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Holly Hargreaves posted on Wednesday, June 05, 2013 - 8:14 am
Good Morning,

May I be directed to resources regarding information on the minimum sample size required for Latent Class Analysis (LCA) as well as information on conducting a power analysis for LCA? I have completed coursework in structural equation modeling and attended Dr. Muthen's training at the 2013 APS Convention however I have only recently began studying LCA and conducting analyses on MPlus therefore any information and/or insight would be greatly appreciated. Thank you very much.

Kind Regards,

Holly
 Bengt O. Muthen posted on Wednesday, June 05, 2013 - 10:19 am
One useful starting point is:

Nylund, K.L., Asparouhov, T., & Muthén, B. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling. A Monte Carlo simulation study. Structural Equation Modeling, 14, 535-569.

Others may have other suggestions.
 Holly Hargreaves posted on Wednesday, June 05, 2013 - 11:25 am
Thank you very much Dr. Muthen. It is my understanding that LCA is not advised with sample sizes smaller than n=100. Is n=117 sufficient in some cases? Any insight as to general rules of thumb would also be greatly appreciated. I sincerely appreciate your time.
 Bengt O. Muthen posted on Wednesday, June 05, 2013 - 1:23 pm
It all depends on how well the classes are separated. With binary outcomes that means how different the thresholds (conditional item probabilities) are. I have done successful mixture modeling with only 30 subjects, so I wouldn't rule out using LCA for 117 subjects. General rules of thumb are not worth much for mixtures because results depend so much on the specifics of your situation. That's why we suggest Monte Carlo studies - see the UG chapter 12. See also the Monte Carlo article:

Muthén, L.K. & Muthén, B.O. (2002). How to use a Monte Carlo study to decide on sample size and determine power. Structural Equation Modeling, 4, 599-620.

with scripts on our website.
 Holly Hargreaves posted on Wednesday, June 05, 2013 - 1:28 pm
Thank you again, I will ensure to read the chapter and article. Is the chapter referenced above on your website as well?
 Linda K. Muthen posted on Wednesday, June 05, 2013 - 1:32 pm
Both items are on the website.
 Silvia Colonna posted on Thursday, November 29, 2018 - 4:18 am
Dear all,

following up this trend, I also have a question about sample size in LPA. I specifically would like to know the power/suitability of my sample size (N=223) to identify different profiles based on 12 variables. I know that usually the rule of thumb is that you have 30 participants for each variable that you're including in the model, is this the case for a LPA as well?
I have read the below mentioned paper but I did not find this specific information.
Are there any resources about this that I can consult?

Thank you for your help.

Kind regards
Silvia
 Bengt O. Muthen posted on Thursday, November 29, 2018 - 10:53 am
I am not familiar with that 30 rule - and I am not convinced that it is very useful since so many factors play into the power, not the least of it being how different the classes are for the means of the outcomes. There are LPA papers on our website under Papers, LCA. If you don't find power studies there, have a look at how we generated data for our UG LPA examples - each UG example has a corresponding Monte Carlo run that generates its data. You find those at

http://www.statmodel.com/ugexcerpts.shtml

Such Monte Carlo runs can be used to get an idea of power.
 Silvia Colonna posted on Friday, November 30, 2018 - 10:28 am
Thank you very much for your quick response, I'll definitely have a look at those.

Kind regards,
Silvia
 shonnslc posted on Friday, June 19, 2020 - 9:10 am
Dear all,

I am doing power analysis for LCA. My aim is to focus on LMR and BLRT. I have read Nylund et al.'s (2007) article but not sure how to do the exact same thing in Mplus since there is no syntax provided there. So, I used Monte Carlo simulation for LCA while asking Tech 11 and Tech 14 in the code:

1. Is this the right way to do power analysis for LCA if my focus is on LMR and BLRT tests?

2. Do I look at the proportion of replications rejected at the 5% level in Tech 11 and Tech 14 section to determine the power?

3. I don't know why when the p-value for LMR test is less than .05, my proportion of replications rejected at the 5% level will always be 1 and when p value is higher than .05, my proportion is always 0 like this:

when p < .05

VLM Likelihood Ratio 1.000
Replications attempted 100
Replications completed 100

when p > .05

VLM Likelihood Ratio 0.000
Replications attempted 100
Replications completed 100

Thank you!!
 Bengt O. Muthen posted on Friday, June 19, 2020 - 6:18 pm
I would not recommend doing this. It is quite advanced and your questions suggest that you are not yet at that level. An easier topic is to look at BIC.

But you can ask on SEMNET.
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: