A0436
Title: Doubly robust identification and estimation of the LATE model with a continuous treatment
Authors: Yingying Dong - University of California Irvine (United States) [presenting]
Abstract: Identification and estimation of the LATE model with a continuous treatment are considered. Two alternative restrictions are discussed on the first-stage instrument effect heterogeneity that allows for causal identification - monotonicity and treatment rank similarity. The former is popular in the LATE literature, while a slightly stronger version of the latter is exploited in the non-separable IV model literature. Neither assumption implies the other. Both assumptions can, at best, be partially tested. Causal estimands with doubly robust properties are proposed in that they are valid under either alternative restrictions.Further semiparametric estimators are proposed, and the asymptotic properties of these estimators are derived. When monotonicity holds, the primary estimand reduces to the standard LATE Wald ratio; otherwise, when treatment rank similarity holds, the approach allows for identifying treatment effect heterogeneity at different (conditional) treatment quantiles. The proposed estimators are applied to evaluate the impacts of neighbourhood poverty rate (a continuous treatment variable) on adults' labour market outcomes using the Moving to Opportunity (MTO) social experiment.
