Causal Inference for Social Network Data.

Area of Science:

• Social network analysis
• Causal inference
• Statistical modeling

Background:

• Observational data from social networks presents unique statistical challenges.
• Previous methods for causal inference in networks had limitations in handling complex dependencies.
• Understanding peer effects requires robust methods that account for network structure.

Purpose of the Study:

• To develop semiparametric estimation and inference methods for causal effects in single social networks.
• To address limitations in existing methods by allowing for multiple sources of dependence.
• To propose novel causal effects relevant to social network interventions.

Main Methods:

• Developed asymptotic results for causal inference with growing dependence on network neighbors.
• Incorporated both information transmission and latent similarities as sources of network dependence.
• Proposed new causal effects tailored for social network structures and interventions.

Main Results:

• The proposed methods allow for complex, growing dependence structures in social network data.
• New causal effects were defined for network-based interventions.
• Reanalysis of obesity peer effects data from the Framingham Heart Study found no evidence of causal peer effects when network structure was accounted for.

Conclusions:

• The developed methods offer a more comprehensive approach to causal inference in social networks.
• Accounting for network structure is crucial when estimating peer effects.
• The findings challenge previous conclusions regarding causal peer effects of obesity.