I have a bunch of
factor variables. I believe the data comes from only a few clusters. I'd like to analyze the data and perform data reduction. I want to know the most important variables for predicting membership in each of the classes.
I've found the
poLCA package. It does almost everything that I need. The
poLCA.posterior function gives the pdf of class membership conditional on variable values. This is OK, but not quite what I want, because there is no data reduction. I don't know which variables I could drop while not sacrificing much in terms of classification accuracy.