Measuring Measurement Error

Abstract

We highlight the importance of measurement error in applied empirical work using
a sample of 2,185 instrumental variable regressions from 326 papers published in top economics and finance journals. If published instruments are valid for measurement error, our estimates imply only 20%-40% of the variance of the average regressor is attributable to the underlying variable of interest. Publication bias does not quantitatively explain our results, although we cannot rule out the influence of heterogeneous treatment effects or instrument invalidity. Our estimator can also bolster identification arguments when IV estimates are unexpectedly large.

Keywords: Measurement error, instrumental variables, endogeneity

JEL Classification: C26, C36