|
Activity Number:
|
357
- Contemporary Multivariate Methods
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #311077
|
|
|
Title:
|
Uplift Modeling for Panel Data Using Switch Doubly Robust Method
|
|
Author(s):
|
Hiroaki Naito* and Hisayuki Hara
|
|
Companies:
|
Doshisha University and Doshisha University
|
|
Keywords:
|
Uplift modeling;
Causal inference;
Machine learning
|
|
Abstract:
|
Uplift modeling is used to optimize effect of intervention by predicting causal effect of treatment for each unit from its covariates. The optimization here means predicting causal effect prior to intervention to select appropriately units that should be treated. Uplift modeling has been developing in marketing science and has been discussed mainly in randomized controlled trial. Recently, uplift modeling is extended also to nonrandomized study where only observational data is available. Athey and Imbens (2015) proposed transformed outcome method (TOM) for nonrandomized study. This method transforms outcome using propensity score. TOM is based on inverse probability weighting (IPW) estimator of the average treatment effect for cross-section data. Saito et al. (2019) proposed switch doubly robust method (SDRM) which is based on doubly robust (DR) estimator and is shown to improve on TOM. In this talk, we extend SDRM to panel data based on DR estimator for difference-in-differences proposed in Li and Li (2019). Under the situation where panel data is available, we compare accuracy of the proposed method with ones in the previous studies by simulations.
|
Authors who are presenting talks have a * after their name.
