Double Robust Estimation with Missing Values

Project Description


Statistical inference for regression models when data are incomplete (item nonresponse) and/or inaccurate (measured with error) is one of the very important issues in empirical research based on nonexperimental data.

Inverse Probability Weighting (IPW) and Multiple Imputation are two major tools to cope with the missing value problem. The recent statistical literature shows that the combination of these two methods yields robustness against misspecification. Despite the appealing properties of the double robust IPW estimator it is not very popular in practice. The stochastic properties of estimation methods using double robust IPW are not yet clearly understood. Therefore, our research project focuses on:

  • Analysis of the asymptotic properties of double robust IPW estimators under different settings (cross-sections, panels, different nonparametric regression models)
  • Generalization of Double Robust Methods (e.g. for the cases where more than one variable is missing)
  • Studies of Small sample properties
  • Possible combination of IPW with matching techniques instead of multiple imputation
  • Applications