I'm hoping for a little guidance on how to model data in the following design. I'm interested in estimating the effect of gender, social context (individual vs group of 3 people), and their interaction on 2 latent variables (male attractiveness & female attractiveness) with 5 indicators each. A colleague suggested using a multilevel design because scores on the 5 indicators are not independent in the group condition. I'm having trouble conceptualizing how to do this with an interaction effect. Are there any examples in the user guide that would be particularly useful for this design?
The multivariate approach used by Mplus takes this type of non-independence into account by modeling it. You do not need to use multilevel analysis.
rgm smeets posted on Tuesday, February 05, 2019 - 11:37 pm
I a ran a LCA on 12,000 patients with 4 classes using 9 indicators (both continuous and categorical) using the MLR estimator with high starts values. I identified a 4-class model based on the lowest BIC, AIC, significant BLRT and high entropy score (above 0.9). Some patients in my dataset can be considered non-independent as they belong to the same family. Do I need to switch to a TYPE=COMPLEX analysis now or can I stick to the TYPE=MIXTURE analysis? And if I switch to TYPE=COMPLEX, is it likely that my model will change or only the SEs?
rgm smeets posted on Wednesday, February 06, 2019 - 6:39 am
I already solved the issue. I ran a sensitivity analysis to check whether the classes would differ if I removed the dependent cases and they did not.