Abstract by Prof. Tim Liao

Testing Coefficient Equality with Bayesian Posterior Information in Generalized Linear Models across Multiple Groups.

Social scientists often need to compare coefficients across groups such as gender, race, and ethnicity. Economists need to test structural change over time. An easy way for testing coefficient equality between groups is to apply the likelihood ratio test in generalized linear models, a method related to the Chow test in linear regression. Such test is about the joint difference in the groups. However, researchers may form hypotheses involving nonnested ones that cannot be handled by conventional tests. This talk employs generalized linear Bayesian modeling by using the posterior information to test parameter equality with information from all relevant models in the model space. As an example, we use a common type of generalized linear model, the logit model, to compare the results based on the LRT statistic and Bayesian posterior information. The examples demonstrate that Bayesian model comparison provides flexibility that is unavailable in conventional tests.

Tuesday - March 11, 1997.
11:00 AM - 2 Illini Hall - PROBABILITY & STATISTICS SEMINAR