Bayesian Full Rank Marginalization for Two-Way Contingency Tables.

摘要: A general approach is proposed for modeling the structure of an r x s contingency table and drawing marginal inferences about all parameters (e.g., interaction effects) in model. The main relevant whenever rs - + 1 5. may also be used to check adequacy Rasch's multiplicative Poisson In general, posterior estimates cell proba- bilities compromise between frequencies fitted values obtained under reduced model spirit Leonard (1975). It possible compute reasonable approximations full densities many inter- est, following (1982) Tierney Kadane (1984). All prior are evaluated with assistance data via a hierarchical Bayes procedure, thus reducing subjectivity involved analysis. An cross classification 5,648 Marine Corps clerical students by school test grade analyzed detail, 96 interactions suggest simplified partitioning collapsing into meaningful 3 2 table.