Intervening on Network Ties.

Area of Science:

• Causal Inference
• Network Science
• Social Network Analysis

Background:

• Emulating randomized control trials via variable interventions is key for causal inference.
• Existing methods primarily focus on independent and identically distributed (iid) data.
• Recent interest has emerged in non-iid settings, particularly networks with interacting agents.

Purpose of the Study:

• To propose and define a novel type of structural intervention, termed network intervention, for analyzing social network contexts.
• To define individual participant and average bystander effects for network interventions.
• To establish identification criteria and theoretical results for network interventions.

Main Methods:

• Defining network interventions as changes to social network structure (creation/severance of ties).
• Developing identification criteria for network interventions.
• Proving theoretical results linking existing identification theory to network interventions via minimally KL-divergent distributions.

Main Results:

• The paper defines individual participant and average bystander effects for network interventions.
• Theoretical results demonstrate that existing identification theory can be applied to network interventions.
• A simulation study validates the estimation of network intervention effects.

Conclusions:

• Network interventions offer a new framework for causal inference in network settings.
• The proposed methods provide a theoretical foundation for analyzing structural changes in social networks.
• This work extends causal inference techniques to complex, non-iid network data.