latent class analysis

Hi all:

I'm interested in using a set of 6 binary indicators to do a latent class analysis and I'm wondering if my strategy can be improved.

We have N=800 schools and I'm interested in developing "profiles" that classify schools on said indicators. I'm also interested in using predictors of latent class membership, most likely in the form of a multinominal logistic regression.

I do have data at the individual (student) level and I plan to aggregate to the school level. For example, if 80% of students in a particular school endorse a given indicator, I will designate that school as scoring a "1" vs "0" on that indicator. Is this the best strategy?

I'm wondering if multilevel extensions of latent class analysis exist and if they do, would they be more accurate methods to use and would they allow me to still form latent classes at the level 2 (school) as opposed to the level 1 (students). The focus of this project is really about schools performance on the indicators.

Why use a LCA specifically, rather than a k-means approach? That way you wouldn't lose any information on the distribution of your indicators. You can still run a MNL (or a discriminant) on the back end.