We have data of 126 persons, with 7 timepoints, where on a continuous outcome, the MADRS score (depression) was measured. To describe the course of the MADRS score over time, we performed an LCA in Mplus (i.e. not mentioning the MIXTURE part in the input). The best fitting result was a 3-cluster model, which seems reasonable and logical. However, LCA seems to be applicable only for cross sectional data (i.e. not taking into account the longitudinal aspect of our data). GMM might be a better option? However, this is where we get confused. We have followed the example in the users guide, chapter 8 (example 8.1), where time scores are fixed. Here we get very different results: all but 5 people have been classified into the second cluster. Do you have an idea why this could be? We are first time users of Mplus, so we are not very experienced... Thanks a lot for your help.
You are looking at two different models -- LCA where growth is not taken into account and GMM where it is. Another model that you might want to consider is LCGA. You would not expect to find the same number of classes with different models. You may find the following papers helpful:
Kreuter, F. & Muthen, B. (2006). Analyzing criminal trajectory profiles: Bridging multilevel and group-based approaches using growth mixture modeling.
Muthén, B. (2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data. In D. Kaplan (ed.), Handbook of quantitative methodology for the social sciences (pp. 345-368). Newbury Park, CA: Sage Publications.
Both of these papers can be downloaded from the website.
Another paper, also available on the website, that you may find helpful is:
Kreuter, F. & Muthen, B. (2007). Longitudinal modeling of population heterogeneity: Methodological challenges to the analysis of empirically derived criminal trajectory profiles. Forthcoming in Hancock, G. R., & Samuelsen, K. M. (Eds.). (2007). Advances in latent variable mixture models. Charlotte, NC: Information Age Publishing, Inc.
Dear Dr & Dr Muthen, I am new to GMM and I would appreciate your help to better understand this paper: Muthen, B. (2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data.
What is the meaning of conditional independence in GMM? What are the differences with conditional indepence in LCA? Any reference about this issue in the context of GMM? Thank you.
LCGA is more like LCA - both have cond'l independence among observed outcomes given class. For GMM and FMM you don't have that because the outcomes correlate also within class due to the (growth) factors. We talk about that in our short course Topic 6. That has references too.
benedetta posted on Wednesday, October 30, 2019 - 9:51 am
I am trying to characterize behavioral paths in adulthood based on binary indicators of smoking measured at 4 points in time, I expect to identify classes such as "never smoked", "quitter" "persistent smoker" etc.. Then I would like to relate such profiles (based on the probability of belonging to a certain class) to socio economic indicators measured at baseline. Would it be appropriate to do so using LCA? or only GMM or LCGA are adequate?
You can use LCA to find the form of the growth curve and then do LCGA or GMM. See our Short Course video and handout for Topic 6, slides 73 and on.
benedetta posted on Sunday, November 03, 2019 - 1:35 am
Thanks a lot! A related question, can I specify the starting values for the thresholds, that I found using LCA, when I move to LCGA or GMM? Fore example %c#1% [Y1$1@-0.729]; [Y2$1@-0.519]; [Y3firstname.lastname@example.org]; [Y4email@example.com]