anonymous posted on Friday, November 02, 2012 - 4:00 pm
I am currently conducting an LCA with 5 ordinal indicators (5-level). I have a few questions: 1. Although there are many 0s (and thus potentially strong floor effects), I also conducted an LPA of these data treating the ordinal indicators as continuous variables. I noticed that the entropy of these models are excellent when indicators are treated continuously in LPA (0.96 to 1.00), but remain fairly low when they are treated as ordinal indicators in LCA (0.67 to 0.59). What is the reason for this? Would the higher entropy levels justify using LPA instead of LCA? 2. Using LCA and treating the indicators as ordinal variables, Iíve tested a 6 class model. However, the output states that only 1 out of 5 classes are replicated for the bootstrapped likelihood ratio test and I need to increase my LRTrandom starts. But, when I increase the LRTrandom starts, there is not enough memory.Is there any way to address this?
You will get different results if you treat a categorical variable as categorical in one analysis and continuous in the other. If you have strong floor effects, you should treat the variable as categorical. Although you might like the results better from the continuous analysis, the nature of the categorical variables is not being taken into account.
See Web Note 14 on the website for further information about LRTSTARTS.
I am currently conducting an LCA. Hereby the following information: Number of groups 1 Number of observations: 88 Number of dependent variables: 4 Number of independent variables: 0 Number of continuous latent variables 0 Number of categorical latent variables 1
Number of missing data patterns 5 Number of y missing data patterns 5 Number of u missing data patterns 0
2 class versus 3 class model AIC, BIC and adjusted BIC for the 3 class model are better (lower) than those of the 2 class model
For 3 class model: Entropy is 0.86 Posterior probabilities: 0.97, 0.99, 0.96 P value of the LMR-LRT is 0.11 P value of the BLRT is 0.000
My questions are: 1. is LCA possible with this sample size (n=88)? 2. How about the 3 class model?