Causal inference, social networks and chain graphs.

Area of Science:

• Causal inference
• Network analysis
• Social science

Background:

• Traditional statistical and causal inference assumes independent individuals.
• Interactions in social networks lead to treatment spillover and outcome contagion.
• Existing models for network data are often data-intensive and high-dimensional.

Purpose of the Study:

• To propose a parsimonious parameterization for network data with interference and contagion.
• To introduce chain graphs as a suitable model for analyzing such data.
• To demonstrate the application of chain graphs for causal inference in social networks.

Main Methods:

• Utilized a parsimonious parameterization corresponding to chain graphs.
• Argued that chain graphs approximate marginal distributions in longitudinal network data.
• Applied chain graphs to observational data from US Supreme Court decisions (1994-2004) and simulations.

Main Results:

• Chain graphs provide a computationally feasible approach for network causal inference.
• The proposed method effectively models interference and contagion in interacting populations.
• Demonstrated successful application in analyzing collective decision-making in social networks.

Conclusions:

• Chain graphs offer a practical and effective framework for causal inference in network settings.
• This approach overcomes limitations of existing high-dimensional and data-intensive models.
• The findings have implications for understanding social influence and collective behavior.