Reinforced Borrowing Framework: Leveraging Auxiliary Data for Individualized Inference.

Area of Science:

• Statistical modeling
• Data science
• Machine learning

Background:

• Auxiliary data is increasingly used to enhance individualized inference.
• Existing methods like multisource exchangeability models (MEM) borrow information from supplemental sources based on parameter exchangeability.
• These methods often ignore other valuable information that could determine source exchangeability.

Purpose of the Study:

• To propose a generalized reinforced borrowing framework (RBF) for enhanced individualized inference.
• To leverage auxiliary data, including a distance-embedded prior, to reinforce inference on target parameters.
• To improve inference accuracy with minimal computational overhead.

Main Methods:

• Developed a generalized reinforced borrowing framework (RBF).
• Incorporated a distance-embedded prior utilizing auxiliary information beyond the target parameter.
• Applied RBF to analyze the impact of the COVID-19 pandemic on individual behaviors.

Main Results:

• RBF achieves 20%-40% lower Mean Squared Error (MSE) compared to existing methods.
• The framework effectively leverages diverse auxiliary information sources.
• Demonstrated improved individualized inference with minimal additional computational burden.

Conclusions:

• The proposed RBF offers a significant improvement over existing methods for individualized inference.
• RBF enhances the utilization of auxiliary data by considering a broader range of information.
• This framework has practical applications, as shown in the COVID-19 behavioral study.