Using observed sequence to orient causal networks.

Area of Science:

• Causal inference
• Network analysis
• Longitudinal data analysis

Background:

• Causal network learning often relies on cross-sectional data, which lacks inherent temporal information.
• Sequence is a critical component of causality, typically observed in longitudinal studies.

Purpose of the Study:

• To develop a method for orienting arc directions in causal networks using longitudinal data sequences.
• To enhance the causal interpretability of network models by integrating temporal event order.

Main Methods:

• Learned network structures from cross-sectional data using established algorithms.
• Determined longitudinal event sequences using Probabilistic Contrast and Goodman and Kruskal error reduction methods.
• Prohibited arc directions in cross-sectional networks that contradicted the learned longitudinal sequences.

Main Results:

• The proposed method significantly reduced arc direction errors in simulated causal networks.
• Increased number of events in the network led to greater improvements in accuracy.
• Real-world data analysis showed increased agreement among learned networks when incorporating longitudinal sequence information.

Conclusions:

• Combining longitudinal sequence information with cross-sectional network learning algorithms is feasible.
• This integrated approach enhances the causal interpretability of network models.
• The method offers a more robust way to infer causal relationships by accounting for the observed sequence of events.