Estimation and testing of multiplicative models for frequency data (Q2338092)

Multiplicative models are a subclass of relational models where the probabilities of certain cells are the product of the probabilities of other cells. In this paper the author provides a new algorithm to improve the Maximum Likelihood Estimation of the parameters for multiplicative models. With the proposed new technique the convergence rate is quadratic rather than linear. Moreover, the theory developed in order to define the new algorithm is used to better understand the geometry of such a class of models. The theory is illustrated by means of a small example of an incomplete \(2^3\) contingency table from Market Basket Analysis.