Targeted minimum loss based estimator that outperforms a given estimator.

Area of Science:

• Statistics
• Causal Inference
• Semiparametric Models

Background:

• Targeted Minimum Loss-based Estimation (TMLE) is a framework for constructing semiparametric, locally efficient, and double-robust estimators.
• Existing TMLE methods are applied to censored data and causal inference models.
• There is a need for TMLEs that are at least as efficient as user-supplied asymptotically linear estimators.

Purpose of the Study:

• To demonstrate the construction of a TMLE that achieves at least the efficiency of a user-supplied asymptotically linear estimator.
• To show that this enhanced TMLE can incorporate empirical efficiency maximization while maintaining double robustness.
• To illustrate the method for estimating the additive average causal effect of a point treatment.

Main Methods:

• Construction of a Targeted Minimum Loss-based Estimator (TMLE) with enhanced efficiency properties.
• Incorporation of empirical efficiency maximization techniques within the TMLE framework.
• Focus on semiparametric censored data and causal inference models, specifically additive average causal effects.

Main Results:

• The developed TMLE is shown to be at least as efficient as user-supplied asymptotically linear estimators.
• The TMLE successfully integrates empirical efficiency maximization, building upon prior work.
• Double robustness is retained in the proposed TMLE construction.

Conclusions:

• A novel TMLE construction is presented that enhances efficiency and retains desirable statistical properties.
• This approach provides a valuable tool for causal inference, particularly for estimating average causal effects.
• The method offers a robust and efficient alternative for semiparametric estimation problems.