Year of Graduation
Multivariate Log-Linear Analysis of Sparse Contingency Tables: Characteristics and Limitations in the Sociological Practice
Applied Methods of Social Analysis of Markets
The main focus of the paper is the problem of the non-existent maximum likelihood estimate (MLE) in loglinear models on sparse datasets, which can lead to an incorrect evaluation of goodness-of-fit tests, the incorrect estimation of the significance of the coefficients, and in some cases to an incorrect estimation of the coefficients themselves. Based on the analytical work, the literature review and series of statistical experiments, I propose recommendations on working with log-linear analysis, that allow minimizing, and in some cases excluding the probability of occurrence of a nonexistent maximum likelihood estimates.