The Model and Methodology

Abstract

After an early period of interest in models containing unobservable variables, the attention of econometric theory switched to models where the stochastic disturbances were only in the form of “shocks” perturbating the structural equations. It is only in recent years that the interest in models with stochastic disturbances affecting the variables (“errors”) has been revived. The evolution of errors-in-variables (EIV) models in the econometric literature is described in Goldberger [1972] [1974] and Griliches [1974]. Part of the unpopularity of EIV models was undoubtedly due to the identification problems that unobservable variables could create.^1/ To formulate identified models, strong a priori information was required. For models with normal likelihoods it seemed that consistent estimators could only be found if some of the variances (or their ratios) were known, if additional instrumental variables were available, or, from a Bayesian approach, if informative prior distrubutions for the parameters were used (see Johnston [1965, chap. 6] and Zellner [1971, chap. V]).^2/ None of this prior information seemed very appealing to econometricians.