Confidence and Conflict in Multivariate Calibration

SUMMARY Multivariate calibration involves using an estimated relationship between a multivariate response Y and an explanatory vector X to predict unknown X in future from further observed responses. In controlled calibration with multivariate normal errors, the profile likelihood function for the unknown X (denoted 4) displays a term which measures the mutual inconsistency of the given response vector (denoted Z) in predicting 4. This inconsistency diagnostic fundamentally differentiates the behaviour of likelihood based and Bayes "confidence" regions from those of the unconditional sampling approach. In addition the diagnostic serves to pinpoint an inadequate response vector Z.