An Efficient Computational Method for Causal Inference in High-Dimensional Data: Neighborhood-Based Cross Fitting

• Activity Number: 32 - Nonparametric Methods with High-Dimensional Data
• Type: Contributed
• Date/Time: Sunday, August 7, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Oluwagbenga David Agboola*
• Companies: University of Northern Colorado
• Keywords: High dimensional data; Causal inference; Data splitting; Post-selection bias; Repeated data splitting; Cross fitting

Abstract

This study proposed a new computationally efficient data splitting method called Neighborhood-Based Cross Fitting (NBCF) for double machine learning in causal inference on high dimensional data, which are increasingly popular in various physical and social disciplines. A common existing approach of repeatedly splitting data was suggested to address the overfitting problem in high dimensional statistics, however it is computationally expensive. The proposed method deals well with the problem of post-selection bias in causal inference in the presence of high dimensional confounders and provides an equivalent performance in unbiased estimation as repeated data splitting to expand the scope of function class by Donsker. Simulation studies were conducted to demonstrate that the proposed neighborhood-based approach is not only more computationally efficient than the existing sample splitting methods but also better in bias reduction compared with other existing methods. Under certain conditions, simulation results further showed that the proposed estimators are asymptotically unbiased and normally distributed, which allows construction of valid confidence intervals.