Across-Regime Covariance Restrictions in Treatment Response Models.

Chib and Hamilton (2000) discuss Bayesian estimation of treatment response models subject to the restriction that the cross-regime correlation parameter is zero. This note points out important consequences of that restriction, and argues that the range of applicability of such an approach is necessarily limited. We also briefly discuss methods for fitting these models which do not impose such a restriction, and as a consequence, are not subject to the described limitations.