Bayesian model comparison in generalized linear models across multiple groups

This paper extends the statistical method known as generalized linear Bayesian modeling developed by Adrian Raftery to the comparison of generalized linear models across multiple groups. The extension considers all relevant hierarchical models in the model space and tests parameter equality across groups by using Bayesian posterior information from the models. The conclusion drawn by using the proposed approach tends to be more conservative than Raftery's method and the conventional likelihood ratio test, as the examples demonstrates. (C) 2002 Elsevier Science B.V. All rights reserved.