Characterizing the Instrumental Variable Identifying Assumption as Sample Selection Conditions

We build on Rosenzweig and Wolpin (2000) and Keane (2010) and show that in order to ful ll the Instrumental variable (IV) identifying moment condition, a policy must be designed so that compliers and non-compliers either have the same average error term, or have an error term ratio equal to their relative share of the population. The former condition (labeled Choice Orthogonality) is essentially a no-selection condition. The latter one, referred to as Weighted Opposite Choices, may be viewed as a distributional (functional form) assumption necessary to match the degree of selectivity between compliers and noncompliers to their relative population proportions. Those conditions form a core of implicit IV assumptions that are present in any empirical applications. They allow the econometrician to gain substantial insight about the validity of a speci c instrument, and they illustrate the link between identi cation and the statistical strength of an instrument. Finally, our characterization may also help designing a policy generating a valid instrument.