Latent Classes and Exponential Families
In an exploratory setting, the latent class model assumes that conditioning on an unobserved class membership leads to conditional independence among the observed variables. The EM algorithm is often used to find MLEs under the latent class model, and a detailed proof of the convergence of the EM algorithm is given. As more general theoretical framework, exponential families of probability distributions are introduced and some of their basic properties are proved. Exponential families not only give a background for many of the concepts and results in this book, e.g., marginal distributions and odds ratios are special cases of fundamental parameters in an exponential family, but also constitute the basis of most of the further developments in the book.
- 37.Haberman, S.J.: Iterative scaling procedures for log-linear models for frequency tables derived by indirect observation. Proc. Amer. Statist. Assoc. (Statist. Comp. Sect., 1975), 45–50 (1976)Google Scholar
- 92.Wu, C.F.J.: On the convergence properties of the EM algorithm. Annals of Statistics 11, 95–103 (1983)Google Scholar