Nonparametric Imputation using Random Forest

Tuesday, February 25, 2020 - 12:30pm to 1:30pm
Event Type: 

Speaker: Hengfang Wang

Abstract: Imputation is a popular technique for handling item nonresponse. Kim and Rao(2009) proposed a linearization method for variance estimation with imputed data when the true model is parametric. In this talk, I’ll talk about nonparametric imputation with kernel-based random forest(KeRF). First of all, the basic concept of general random forest(Breiman 2001) and KeRF will be introduced. Then the connection between kernel smoothing and KeRF will be discussed. I’ll then propose an intuitive variance estimator based on a linearization method. Finally, simulation results are presented under different settings and discussion about difficulties in variance estimation under random forest imputation is illustrated.