Measuring Measurement Error

Abstract We collect 2,848 instrumental variable (IV) regressions from 391 papers published in top economics and finance journals and show that estimated IV coefficients are often much larger than their OLS counterparts, even when theory suggests they should be smaller. We derive meta-regression estimators that distinguish measurement error bias from omitted variable bias and simultaneity bias; applying our estimators to a set of 1,040 comparable regressions from 139 papers, we find that measurement error shrinks reported coefficients by about two to four times. Measurement error remains quantitatively important even after accounting for weak instruments, publication bias, instrument invalidity, and heterogeneous treatment effects. Our results suggest that measurement error is a severe, pervasive, and understated source of bias in economics and finance.