Development Economics X Paper Model Twelve
We share a new method for performing instrumental variables and the local average treatment effect (LATE) on distribution functions. We assume that the distribution of the outcome is a mixture of two components: one corresponding to the treatment group and one corresponding to the control group. We use an instrumental variable that satisfies the standard assumptions of independence, exclusion, and monotonicity to identify and estimate the mixing proportion, which is equivalent to the proportion of compliers. We then estimate the LATE as the Wasserstein distance between the two components of the mixture model. Our method provides a flexible and robust way to quantify causal effects on distribution functions in settings where there is unobserved confounding or non-compliance.
Opoku-Agyemang, Kweku (2023). "Distributional Instrumental Variables: Identification and Estimation." Development Economics X Paper Model Twelve.
© 2023 Development Economics X Corporation. All Rights Reserved.