Person-centered treatment (PeT) effects: Individualized treatment effects using instrumental variables

摘要

I describe a command, petiv, that uses a local instrumental-variables (LIV) approach to estimate person-centered treatment effects for a variety of specifications for the LIV estimand as outlined in Basu (2014, Journal of Applied Econometrics 29: 671-691). The petiv command creates a new variable in the dataset that contains the person-centered treatment effects for each individual in the dataset. However, the command takes the validity of the instrumental variables and the specification of the LIV estimand as given. Appropriateness of these features of an LIV analysis should be determined before running the petiv command. The individual effects can be used to answer distributional questions and can also be easily aggregated to obtain mean treatment-effects estimates.