Abstract This paper considers the use of multiple proxy measures for an unobserved variable and contrasts the approach taken in the measurement error literature to that of the model specification literature. We find that including all available proxy variables in the regression minimizes the...