Most models encountered in areas like contingency table analysis and multivariate statistics are defined in terms of polynomial constraints or polynomial parametrizations. In particular, classical latent variable models such as factor analysis, instrumental variable and latent class models are of this type. Latent variable models typically have geometrically irregular parameter spaces that can lead to non-standard behavior of statistical procedures. Focusing on hypothesis tests, this talk will discuss the connections between asymptotic distribution theory and algebraic model geometry.
Keywords: Algebraic statistics; Likelihood ratio test; Wald test; Large sample theory
Biography: Mathias Drton is an Associate Professor at the University of Chicago.