Directed Cyclic Graphs for Simultaneous Discovery of Time-Lagged and Instantaneous Causality from Longitudinal Data...

Area of Science:

• Causal Inference
• Statistical Learning
• Biostatistics

Background:

• Longitudinal observational data presents challenges for causal discovery due to complex temporal dependencies.
• Existing methods often struggle to identify both instantaneous cyclic and time-lagged causal structures simultaneously.

Purpose of the Study:

• To develop a novel causal discovery framework for longitudinal data.
• To achieve unique identifiability of directed graphs with general cyclic patterns.
• To simultaneously discover time-lagged and instantaneous causality.

Main Methods:

• A novel framework integrating instrumental information from longitudinal data.
• Development of a causal identification theory for directed acyclic graphs (DAGs) with general cyclic patterns.
• Fully Bayesian structural learning approach.

Main Results:

• The proposed model demonstrates general identifiability under common causal discovery assumptions.
• Achieved unique causal identifiability for directed graphs with general cyclic patterns, a novel theoretical contribution.
• Outperformed state-of-the-art methods in extensive simulations and a real-world application.

Conclusions:

• The developed framework provides a robust method for causal discovery from longitudinal data.
• The model successfully identifies complex causal structures, including cyclic dependencies.
• The approach offers superior utility and identifiability compared to existing causal discovery techniques.