Model Testing for High-Dimensional Contingency Tables with Application in Genetics

Abstract

In the paper it is supposed that random sequences is a finite-order Markov chain. The analysis of high order interactions in such sequences leads to large and sparse contingency tables. A special data presentation form and data transformation via multidimensional scaling enables us to apply generalized logit model and reduce the problem to the fitting nonparametric multinomial logistic regression with a few quantitative variables.