Constructing targeted minimum loss/maximum likelihood estimators: a simple illustration to build intuition

Area of Science:

• Causal Inference
• Statistical Learning
• Machine Learning Applications

Background:

• Machine learning is frequently employed to estimate nuisance functions in causal inference.
• The efficient influence function (EIF) provides a robust framework for integrating machine learning into causal inference methods.
• Valid statistical inference, such as estimating confidence intervals, is crucial for reliable causal effect estimation.

Purpose of the Study:

• To provide an accessible guide on constructing targeted maximum likelihood/minimum loss estimators (TMLE) using the EIF.
• To bridge the gap between advanced statistical literature on EIF and applied researchers' needs.
• To demonstrate the practical application of EIF in machine learning-driven causal inference.

Main Methods:

• Illustrating the construction of TMLE from the EIF.
• Leveraging machine learning algorithms for nuisance parameter estimation within the EIF framework.
• Focusing on practical implementation for applied researchers.

Main Results:

• Demonstrated a clear pathway to constructing TMLE from EIF.
• Showcased how to incorporate machine learning for enhanced estimation.
• Provided a tutorial format for easier understanding and application.

Conclusions:

• TMLE construction from EIF is a valuable technique for applied causal inference.
• This approach facilitates valid inference, including confidence intervals, when using machine learning.
• The tutorial aims to make complex statistical methods more accessible to a broader research audience.