Identifiability of Bivariate Mixtures: An Application to Infant Mortality Models

Eric Frimpong, University at Albany, State University of New York (SUNY)
Timothy B. Gage, University at Albany, State University of New York (SUNY)
Howard Stratton, University at Albany, State University of New York (SUNY)

Identifiability of the parameters for a Mixture of Bivariate Densities (MBD) in the form f(x,y;φ,θ,π)=πf(y|x;φ1)f(x;θ1)+ (1-π)f(y|x;φ2)f(x;θ2) is considered. Particular attention is given to θ1≠θ2 (i.e. marginal of x is a nondegenerate mixture). Characterzations of identifiability that includes extensions of Hennig (2000), Hunter (2007) and Gage (2004) models are given. These identified models are applied to characterize latent subpopulations related to infant mortality and survival in several NCHS linked Birth/Death data sets.