A Structural Errors-in-Variables Model with Heteroscedastic Measurement Errors under Heavy-Tailed Distributions

Abstract

Errors-in-variables (measurement error) models are important issues in statistics and widely used in chemistry, physics, econometrics and medical sciences, etc. In this working paper, we discuss point estimation of the parameters in a structural errors-in-variables model with heteroscedastic measurement errors, when the observations jointly follow scale mixtures of normal distributions. The model with and without equation error are both included in our discussion. Compared with the method-of-moments estimators, maximum likelihood estimates are discussed through the EM iterative algorithms.