Reem Saeed posted on Thursday, November 07, 2013 - 2:08 pm
I have done an EFA on a set of continous variables, and have attempted to replicate these using latent profile analysis. However, the factor loadings for the variables are somewhat different than the class membership for the latent profile analysis. So, in other words, the interpretation of the latent structure differs slightly between the 2 methods when both are applied using the exact same dataset.
Is this appropriate? And how would I explain such differences? If I were put in a situation where eigenvalues in an EFA were above 1 for two variables, and the LFA model for say a 3 or 4 class model was a good fit. Which one should I choose and why?
Xu, Man posted on Friday, March 18, 2016 - 7:06 am
Thanks! That is very handy. I ran a quick analysis and saw different factor loading patterns in different classes (for the same number of factors). I guess this is where population heterogeneity comes in such that different sub populations have different factor structures.
However, is it possible to constrain the loadings of the factors, and if so, are the classes based on factor means the same way a LPA of CFA factors?