Dear group - I need some help with a longitudinal applications of LCA (details below). Thanks!
(1) My analytic approach: I have a two-wave longitudinal dataset, and I want to assess the latent pathways of beliefs simultaneously. The dataset consists of 5 binary items, duplicated over two waves of the survey, with the same individuals answering each question.
(2) With these data I want to fit a multilevel LCA that classifies the latent pathways of belief patterns from Wave 1 to Wave 2. The result will be latent classes of beliefs and their change over time.
In any case, I've read a few studies but I'm not even sure if this model describes a GMM, a Latent Growth Curve LCA, or a plain LCA with all 10 items dumped in. I.e., it is unclear whether items were estimated within Wave (1 or 2), within individuals (participants surveyed at each time) or none of the above (ignored and treated as a cross-sectional analysis).
Long story short: (1) Within Mplus, what analytic approach captures this analysis? (I know it is not LTA, as I do not want to examine the transition probabilities; I only want to examine the holistic latent class pathways from wave 1 to Wave 2)
(2) What are some syntax examples to help me through this process?
Dear Dr. Muthen - Thank you for your query; it's my question as well. I've read several papers that never reference LTA. These papers describe latent "pathways" and will sometimes describe them as "multilevel latent class models," but in several papers I see no reference to LTA or Markov models.
In some papers it appears the authors may even be estimating all items in a panel simultaneously but one recent review calls this "multilevel latent class analysis." Accordingly, I'm not sure how to write the Mplus syntax.
In these and related papers authors are not discussing transition probabilities. Rather they want to unmix the entire sample into different latent pathways. This is what I meant by "holistically."
In a follow-up post are two examples from the literature that I hope clarifies my own confusion over the language sometimes used and of the approach to take.
I'd love to know whether (1) these are indeed distinct from LTA (as it sounds to me); and (2) if "multilevel LCA" is distinct, how one might model this in Mplus.
Here are two examples in the life course (transition to adulthood) literature in particular. Thanks! :-) - Nate
Example 1: Below is an example that appears to model panel data using multi-group LCA, but they make no mention of transition probabilities or of latent changes over time. If it is a multilevel LCA, I would assume they would index the items using some grouping identifier, although I can't tell if this would be items nested within individuals over the panels or if it would be an index group items. Or perhaps neither. They don't say.
Oesterle, Sabrina, J. David Hawkins, Karl G. Hill, and Jennifer A. Bailey. “Men’s and Women’s Pathways to Adulthood and Their Adolescent Precursors.” Journal of Marriage and Family 72, no. 5 (October 1, 2010): 1436–53. doi:10.1111/j.1741-3737.2010.00775.x.
Example 2: This model explicitly describes "multilevel latent class models" with the posterior probabilities conditional (if I get it) for each panel of the longitudinal study. I don't know if this is the same as LTA, especially given their language. Perhaps its a difference in communication based on culture and software; I'd love to know! :-)
Vuolo, Mike, Jeylan T. Mortimer, and Jeremy Staff. “Adolescent Precursors of Pathways From School to Work.” Journal of Research on Adolescence 24, no. 1 (March 1, 2014): 145–62. doi:10.1111/jora.12038.
The article by Oesterle et al uses Mplus LCA is some fashion so why don't you ask them how they set up the analyses.
The article by Vuolo et al uses a second-order LCA which I am not familiar with but should also be possible in Mplus. It looks like it has a latent class variable behind a set of observed indicators at each time point and then uses those T latent class variables as indicators of a second-order latent class variable. The second-order latent class variable therefore has T nominal indicators in line with UG ex 7.8, although the nominal indicators are latent.