A simple improvement of the IV estimator for the classical errors-in-variables problem

Two measures of an error-ridden explanatory variable make it possible to solve the classical errors-in-variable problem by using one measure as an instrument for the other. It is well known that a second IV estimate can be obtained by reversing the roles of the two measures. We explore a simple estimator that is the linear combination of these two estimates, that minimizes the asymptotic mean squared error. In a Monte Carlo study we show that the gain in precision is significant compared to using only one of the original IV estimates. The proposed estimator also compares well with full information maximum likelihood under normality.