Title: DNN: A two-scale distributional tale of causal inference
Authors: Yingying Fan - University of Southern California (United States)
Jingbo Wang - University of Southern California (United States)
Jinchi Lv - University of Southern California (United States) [presenting]
Abstract: The problem of heterogeneous treatment effect estimation and inference in nonparametric regression models is considered. To reduce the bias in the k-nearest neighbors estimation method, we exploit the idea of subsampling. Then, by comparing and contrasting the k-nearest neighbors estimators with different subsampling scales, we are able to successfully achieve desired bias reduction. Under some mild regularity conditions, the resulting new DNN estimator is proved to be asymptotically unbiased and have asymptotically normal distribution. The method and theoretical results are supported by several simulation examples. The approach is also applied to a child's birth weight data set to study the heterogeneous treatment effect of smoking.