Bayesian analysis of autocorrelated ordered categorical data for industrial quality monitoring

Presently available methods to analyze the link between explanatory variabl es and all ordered categorical response implicitly assume independence. Thi s hypothesis is no longer valid when data ore collected over time. We assum e temporal dependence and introduce autocorrelation using a latent-variable formulation. Due to intractable distributions, we resort to Gibbs sampling for statistical inference within the Bayesian paradigm. Variable selection is also addressed and appears as a straightforward by-product of this fram ework. We illustrate the method by analyzing on-line quality data that poss ess such autocorrelation.