Estimating a system of linear equations with survey data - ARCHIVED

This paper develops a framework for estimating a system of linear equations with survey data. Pure design-based sample survey theory makes little sense in this context, but some of the techniques developed under this theory can be incorporated into robust model-based estimation strategies. Variance estimators with the form of the single equation “linearization” estimator are nearly unbiased under many complex error structures. Moreover, the inclusion of sampling weights in regression estimation can protect against the possibility of missing regressors. In some situations, however, the existence of missing regressors can make the estimation of a system of equations ambiguous.