Estimation of linear errors-in-variable models with error free covariates: a backfitting approach

Abstract

We present a backfitting algorithm to estimate linear regression mod- els having both error-prone and error-free covariates as predictors. The algorithm assumes that the variance-ratios are known, and it is particulary efficient when several explanatory variables are included. The resulting estimators are shown to be unbiased and to perform well as compared to method-of-moments estimators which are usually employed when the variance ratio is known.