Navigated Weighting to Improve Inverse Probability Weighting for Missing Data Problems and Causal Inference

22 May 2020  ·  Hiroto Katsumata ·

The inverse probability weighting (IPW) is broadly utilized to address missing data problems including causal inference but may suffer from large variances and biases due to propensity score model misspecification. To solve these problems, I propose an estimation method called the navigated weighting (NAWT), which utilizes estimating equations suitable for a specific pre-specified parameter of interest (e.g., the average treatment effects on the treated). Since these pre-specified parameters determine the relative importance of each unit as a function of propensity scores, the NAWT prioritizes important units in the propensity score estimation to improve efficiency and robustness to model misspecification. I investigate its large-sample properties and demonstrate its finite sample improvements through simulation studies and an empirical example. An R package nawtilus which implements the NAWT is developed and available from the Comprehensive R Archive Network (http://cran.r-project.org/package=nawtilus).

PDF Abstract

Datasets


  Add Datasets introduced or used in this paper