The Bad, the Weak, and the Ugly: Avoiding the Pitfalls of Instrumental Variables Estimation

Abstract

Instrumental variables estimation can, in principle, avoid biases that ordinary least squares estimation suffers when explanatory variables are correlated with the disturbances. Finding appropriate instruments is a challenge. This paper uses seven recently published empirical papers to illustrate exemplary practices in IV estimation. Nine strategies for avoiding bad instruments (those correlated with the disturbances), as well as recently developed best practices for coping with weak instruments (those little correlated with the troublesome explanatory variable), are summarized and illustrated. The ugly interpretive perils posed by heterogeneity in agents' behavioral responses to a troublesome explanator are also described and illustrated. All procedures recommended in the paper can be implemented using existing commands (some of them quite newly constructed) for one or more standard econometric packages.

Keywords: instrumental variables, two stage least squares, weak instruments

JEL Classification: C10,C11,C12,C14,C20,C21,C22,C23,C24,C30,C31,C32