Recently during my own research, I found some useful information for people who are doing similar research using LCA: 1. When saving the new dataset including conditional probability and class membership, if you would like to save other variables that are irrelevant to the LCA, such as an ID variable, under the Variable command, we can add one line: auxiliary = id; This will tell Mplus in the new dataset please save the ID variable. In the new dataset, the order of the variables are: variables used to predict class, ID, conditional probabilities, class membership. 2. Another useful book to refer to for doing LCA: Data Analysis with Mplus by Christian Geiser, Chapter 6 3. Multilevel LCA is introduced by a paper by Kimberly L. Henry and Bengt Muthen with the title "Multilevel latent class analysis: An application of adolescent smoking typologies with individual and contextual predictors.
Thanks for sharing! As you said in the lecture that LCA is similar to factor analysis, but is person-centered. I am wondering does an LCA have to be one-factor structure? To be more specific, doing a factor analysis a model can have multiple factors, but it seems that LCA only generates one factor, in which there can be 2, 3, 4, or 5 values (as the example you showed in the lecture).
Not to my knowledge. In LCA we use all indicators or items to identify the 2 + latent classes/groups. However, you might want to check out cognitive diagnostic modeling techniques (see works by Jonathan Templin and his colleagues). You might also find mixture IRT of interest (see a recent how to on Mixture IRT by Ralph De Ayala).
Tremendously helpful. Thank you for sharing knowledge. Such a strong motivation for me that she was also a candidate when she was giving this great presentation!
Thank you so much for the video! It was a great help. I have a sample of 455 families and I wanted to build a latent class regarding the quality of the young-mother relationship informed by both perspectives. Given the sample size, is it advisable to consider another type of analysis? Let me know your thoughts about that. Thanks
I'm so glad that you found this video of great help. Regards to your question, it depends on a lot of variables. Your sample size for LCA is not quite that large but it might be amendable to using LCA. I recommend you read The study titled effect size, statistical power and sample size requirements for the bootstrap likelihood ratio test in latent class analysis. The paper was published in structural equation modeling, 2014, volume 21 issue 4. Pages 534 to 552. I would check out table 8 in particular.
Thank you for this super clear lecture -- I am including both continuous and categorical variables in my analyses. My outcome show that 3 classes suit the data best (and seem theoretically feasible) but I don't understand how to test if the mean differences across latent classes are significant. I understand that the "Model test" function is suppost to test this using a wald statistic but I am not entirely sure how to set up the syntax. Do you happen to have a tip?
This was a really easy to follow presentation. Thank you! Would it be possible to get this or a similar data set to practice? At the moment, I can only use the demo version of Mplus, which can't take more than 6 variables and all the example data sets that I've found online had more than 6.
@@ltolandky Thanks! just wondering, is there any tutorial for RMLCA and LTA? Heard that RMLCA syntax is almost the same as LCA syntax. LTA on the otherhand is time consuming. Can it predict the group too and giving the output save file like how the lca analysis did in MPLUS?