Concerning the problem of estimating structural parameters and smoothing cells in complex contingency tables, we delineate a non standard approach – the "Constrained Fixed Point" - which is pseudo-Bayes (Agresti, 1990, pp. 466) and shares some basic ideas with the Prior Feedback Setup (Casella & Robert, 2002). Applied to a reference data set, it elicits a data-dependent prior which is structured according to the order-restricted log multiplicative model. It might be a practicable alternative to the full maximum likelihood (Ritov & Gilula, 1991) and intensive inferential tools as Gibbs sampling (Brink and Smith, 1996).
File in questo prodotto:
Non ci sono file associati a questo prodotto.