Instrumental Variable Estimation of Dynamic Treatment Effects on a Duration Outcome

This article considers identification and estimation of the causal effect time Z until a subject is treated on duration T. The time-to-treatment not randomly assigned, T right censored by random variable C, T∧C. endogeneity issue using an instrumental explaining independent error term model. We study in fully nonparametric framework. show that our specification generates integral equation, which regression function interest solution. provide conditions rely this equation. assume follows parametric model for purposes. propose procedure give under estimator asymptotically normal. estimators exhibit good finite sample properties simulations. Our methodology applied to evaluate timing therapy burnout.