A Markov chain sampler for contingency table exact inference (Q2488380)

Let \(T\) be a contingency table, let \(g(T)\) be a statistic and let \(B\) be a condition on the margins of \(T\). The paper is devoted to evaluation of \(E(g(T)\mid B)\) via Markov chain Monte Carlo techniques. A Metropolis-Hastings type algorithm is proposed in which some sub-table of \(T\) is changed at each step. An adaptive version of the algorithm for sparse tables is discussed. An application to the exact test of the Hardy-Weinberg equilibrium for genetic data is described.