Thank you for your very useful video ! Considering the criteria we use to determine the right number of classes in a LCA, I was concerned by some inconsistent results in my analysis. I am running a complex mixture model to conduct a LCA and to take into account the fact that I am using complex survey data, and although the plots of AIC, BIC, aBIC, AICC values of each of my models suggest me a two-class model, the vuong-lo-mendell-rubin likelihood ratio suggests me a three class model (p
It is not uncommon for different indices to show divergent results. According to the simulation study by Nylund et al. (2007), the BIC and the bootstrap likelihood ratio difference test (BLRT) tend to perform best in terms of determining the number of classes. Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural equation modeling: A multidisciplinary Journal, 14(4), 535-569. Best, Christian Geiser
Thank you for this very excellent guide. As you suggest, I have labeled one variable that I have that is categorical but does not have ordered categories as nominal in the syntax. Yet, I don't really understand the output that I get here. For the ordered categorical variables I get results in probability scale which are easy to interpret but for this variable I get 'means'. Is it better to also list this variable as a categorical variable and not a nominal variable? Thank you in advance for your reply.
Thank for your informative tutorials. I have a question. I'm using latent transformation analysis. I need help regarding whether the LTA can use a mediating variable. to run Latent Transistion Analysis.
You can use multiple latent class and/or observed grouping ("knownclass") variables in mixture analysis in Mplus. That way, you could have indirect effects between latent class and/or observed grouping variables if that is what you mean. (Already in a standard LTA model with 3 time points the T1 latent class variable has an indirect/mediated effect on the T3 class variable via the T2 class variable.) Best, Christian Geiser
I conducted a latent class analysis, but I am not sure about the result. The probabilities of some classes were 1.000 and the p-value is also 1.00. I want to figure out the problem. Would you please advise?
Perhaps you ran into some boundary parameter estimates. This may or may not indicate a problem. See my video here: ua-cam.com/video/iQb2x5EEBj4/v-deo.html Best, Christian Geiser
You can purchase the latest version from the Mplus homepage at www.statmodel.com/orderonline/categories.php?category=Order-Mplus-Software They also offer a free demo version: statmodel.com/demo.shtml The demo version is only limited by the number of variables that can be used in the analysis. Best, Christian Geiser
The latent class indicator variables can have more than two categories (e.g., ordinal variables). You can even use continuous (metrical) response variables as latent class indicators. For continuous indicators, the analysis is often referred to as latent profile analysis (I have several videos on LPA on this channel). Best, Christian Geiser
@@QuantFish Thanks for the response! Also, is it possible to look at output for several classes at once and to suppress output for just the model fit statistics so we can decide on how many classes to use?
Excellent video! Thanks for the clear clarification, really helpful
Glad to hear it was helpful and thanks for watching!
Hi, I am struggling to understand the output of the LCA, do you have a video explaining? If not, I can come up with a more specific question
Thank you so much
Thank you for your very useful video ! Considering the criteria we use to determine the right number of classes in a LCA, I was concerned by some inconsistent results in my analysis.
I am running a complex mixture model to conduct a LCA and to take into account the fact that I am using complex survey data, and although the plots of AIC, BIC, aBIC, AICC values of each of my models suggest me a two-class model, the vuong-lo-mendell-rubin likelihood ratio suggests me a three class model (p
It is not uncommon for different indices to show divergent results. According to the simulation study by Nylund et al. (2007), the BIC and the bootstrap likelihood ratio difference test (BLRT) tend to perform best in terms of determining the number of classes.
Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural equation modeling: A multidisciplinary Journal, 14(4), 535-569.
Best, Christian Geiser
@@QuantFish Thank you very much !
I dont get plot from my mplus (ver 8.6). It says that it is not provided. Do you know why that happens? Do I need another version?
Not sure why this is. You should be able to get a plot with that version. I would ask the Mplus team about this.
Best, Christian Geiser
Thank you for this very excellent guide. As you suggest, I have labeled one variable that I have that is categorical but does not have ordered categories as nominal in the syntax. Yet, I don't really understand the output that I get here. For the ordered categorical variables I get results in probability scale which are easy to interpret but for this variable I get 'means'. Is it better to also list this variable as a categorical variable and not a nominal variable? Thank you in advance for your reply.
If the variable is dichotomous (binary), I would define it as CATEGORICAL.
Best, Christian Geiser
@@QuantFish Dear Christian, the variable however is not binary, it has four categories. Thank you for your reply. Lara
Thank for your informative tutorials.
I have a question. I'm using latent transformation analysis. I need help regarding whether the LTA can use a mediating variable. to run Latent Transistion Analysis.
You can use multiple latent class and/or observed grouping ("knownclass") variables in mixture analysis in Mplus. That way, you could have indirect effects between latent class and/or observed grouping variables if that is what you mean. (Already in a standard LTA model with 3 time points the T1 latent class variable has an indirect/mediated effect on the T3 class variable via the T2 class variable.) Best, Christian Geiser
I conducted a latent class analysis, but I am not sure about the result. The probabilities of some classes were 1.000 and the p-value is also 1.00. I want to figure out the problem. Would you please advise?
Perhaps you ran into some boundary parameter estimates. This may or may not indicate a problem. See my video here: ua-cam.com/video/iQb2x5EEBj4/v-deo.html
Best, Christian Geiser
Could you please share a link foe the latest version of Mplus
You can purchase the latest version from the Mplus homepage at www.statmodel.com/orderonline/categories.php?category=Order-Mplus-Software
They also offer a free demo version:
statmodel.com/demo.shtml
The demo version is only limited by the number of variables that can be used in the analysis.
Best, Christian Geiser
do the variables have to have only 2 response categories? What if there are more than 2 categories within a latent variable (for example: age=34)
The latent class indicator variables can have more than two categories (e.g., ordinal variables). You can even use continuous (metrical) response variables as latent class indicators. For continuous indicators, the analysis is often referred to as latent profile analysis (I have several videos on LPA on this channel). Best, Christian Geiser
@@QuantFish Thanks for the response! Also, is it possible to look at output for several classes at once and to suppress output for just the model fit statistics so we can decide on how many classes to use?