View article

In applied econometric work in which models are fitted to micro-data, that is to data recording the activity of individual economic agents, it is often the case that data are censored or grouped-or are in some way discrete, and then we generally find parametric models being employed, estimation proceeding using maximum likeli hood methods. I In single equation models it is common to find the normal linear regression model employed to describe the distribution of the latent variate whose grouped or censored realisations are the subject of analy sis and in multiple equation models the assumption of conditional multivariate normality is almost universal.

Normality assumptions are of course very convenient, particularly in multivariate models for which it is hard to find alternative joint distributions possessing desirable properties, for example an unrestricted correlation struc ture and linearity of regressions. However, in …