A Note on Debiased/Double Machine Learning Logistic Partially Linear Model

TL;DR

This paper develops a debiased double machine learning approach for logistic partially linear models, enabling robust inference with high-dimensional or machine learning-based nuisance models, and compares it with existing methods like debiased LASSO.

Contribution

It introduces a new debiased double machine learning estimator for logistic partially linear models that works with high-dimensional and machine learning nuisance models, maintaining double robustness.

Findings

The proposed method achieves rate double robustness.

It can incorporate any blackbox machine learning method.

Comparison with debiased LASSO highlights advantages.

Abstract

It is of particular interests in many application fields to draw doubly robust inference of a logistic partially linear model with the predictor specified as combination of a targeted low dimensional linear parametric function and a nuisance nonparametric function. In recent, Tan (2019) proposed a simple and flexible doubly robust estimator for this purpose. They introduced the two nuisance models, i.e. nonparametric component in the logistic model and conditional mean of the exposure covariates given the other covariates and fixed response, and specified them as fixed dimensional parametric models. Their framework could be potentially extended to machine learning or high dimensional nuisance modelling exploited recently, e.g. in Chernozhukovet al. (2018a,b) and Smucler et al. (2019); Tan (2020). Motivated by this, we derive the debiased/double machine learning logistic partially linear model in this note. For construction of the nuisance models, we separately consider the use of high dimensional sparse parametric models and general machine learning methods. By deriving certain moment equations to calibrate the first order bias of the nuisance models, we preserve a model double robustness property on high dimensional ultra-sparse nuisance models. We also discuss and compare the underlying assumption of our method with debiased LASSO (Van deGeer et al., 2014). To implement the machine learning proposal, we design a full model refitting procedure that allows the use of any blackbox conditional mean estimation method in our framework. Under the machine learning setting, our method is rate doubly robust in a similar sense as Chernozhukov et al. (2018a).