До відкриття латентного бінарного фактора в статистичних даних категорного типу

Abstract

For a discrete model with tree-like structure, we demonstrate that if a separating variable (a root vertex) is binary, then constraints (like “tetrad difference”) hold. Specifically, when there are four or three manifest variables in a model, the “ditetrad-constraint” or “triad-constraint”, respectively, applies. So, these constraints facilitate the discovery of a hidden binary variable (latent class) which is responsible for associations among discrete manifest variables.