A BAYESIAN ANALYSIS OF COVARIANCE FOR COMPARING CHANGES

Abstract

A Bayesian approach to analysis of covariance is proposed, in which it is assumed that the relevant parameters are exchangeable in the sense of De Finetti, i.e., they still have the same prior joint distribution when they are permuted. Actual data analysis shows that this exchangeability assumption results in makred differences in the posterior distributions of effect parameters from analysis using noninformative prior distributions, and hence from the sampling theory results.