CCPR Seminar

"Beyond LATE: Economic Choices and the Identification of Multiple Treatment Effects "

Abstract: “Monotonicity” refers to a condition in choice models with instrumental variables in which a local variation of an instrument shifts all agents toward or against a choice. This paper presents a useful framework to investigate the role of monotonicity in the identification of causal effects in multiple choice models with categorical instrumental variables.  I first examine a new monotonicity condition that applies to unordered choice models with multiple treatments.  Like its analogous property in the binary choice model, I show that unordered monotonicity imply and is implied by additive separability in observables and unobservables in choice equations.  I show that unordered monotonicity may arise from preference properties of choice behavior. I then exemplify the use of preference properties to identify causal effects in choice models where monotonicity does not hold. I show that identification and equivalence results flow from simple properties of binary matrices.