Dr. Adrian Dobra

Department of Statistics and Nursing

University of Washington

Title: " The Mode Oriented Stochastic Search Algorithm for Multi-way Contingency Tables "

We develop a novel stochastic search algorithm for model determination
in multi-way tables of counts. We show how the inherent sparsity of
high-dimensional categorical data is properly addressed through the
conjugate priors for log-linear parameters introduced in Massam and
Liu, 2006. We contrast our approach with related Markov chain Monte Carlo
methods (e.g., Dellaportas and Forster, 1999). We discuss how to perform
the search in the space of decomposable graphical log-linear models,
graphical log-linear models and hierarchical log-linear models. We also
introduce a disjoint clique search that allows us to analyze a
sixteen-dimensional sparse contingency table. Other examples involve
six and an eight dimensional dichotomous tables.