ABSTRACT
Daniel Heinz Department of Mathematics and Statistics, Loyola University of Maryland, Baltimore, MD 21210, USA
One issue with parametric latent class models, regardless of whether or not they feature mixed memberships, is the need to specify a bounded number of classes a priori. By contrast, nonparametric models use an unbounded number of classes, of which some random number are observed in the data. In this way, nonparametric models provide a method to infer the correct number of classes based on the number of observations and their similarity.
