Causal inference and the data-fusion problem.

Area of Science:

• Causal Inference
• Big Data Analytics
• Statistical Methodology

Background:

• Current causal analysis methods face challenges with big data.
• Integrating multiple heterogeneous datasets (data fusion) is complex due to biases.
• Existing bias correction methods are often isolated and model-specific.

Purpose of the Study:

• To unify approaches in causal analysis for big data.
• To develop a general framework for handling biases in heterogeneous data.
• To provide a theoretical solution for data fusion in causal inference.

Main Methods:

• Review of causal analysis concepts and principles.
• Development of a general, nonparametric framework.
• Addressing confounding, sampling selection, and cross-population biases.

Main Results:

• A unified approach to causal analysis in big data environments.
• A nonparametric framework for managing biases from heterogeneous data sources.
• A theoretical solution for data fusion in causal inference tasks.

Conclusions:

• The proposed framework enables valid causal inference from combined heterogeneous datasets.
• This work advances causal inference methodology for big data challenges.
• It provides a robust approach to data fusion, overcoming limitations of individual data sources.