Michele De Nadai (The University of New South Wales)

The use of instrumental variables to identify causal effects is widespread in economics. Justifying the validity of exclusion restrictions and monotonicity in just-identified settings is usually solely based on subjective economic reasoning. We propose a general test for instrument validity for settings with one instrument and one treatment variable that is applicable to a wide variety of frameworks. The test statistic is based on an extension of testable implications first derived by Balke and Pearl (1997) to multivalued treatments. We apply the proposed testing procedure to test for the validity of the same-sex instrument in Angrist and Evans (1998), and we find strong evidence against its validity in identifying the effect of one additional child on labour outcomes.