Joint mixed-effects models for causal inference in clustered network-based observational studies.

Area of Science:

• Social network analysis
• Causal inference
• Bayesian statistics

Background:

• Causal inference in social networks is complex due to network interference.
• Standard methods assume no unmeasured confounding, which is often violated in multilevel network data.
• Latent cluster-level factors can bias exposure and outcome assessments.

Purpose of the Study:

• To develop a Bayesian inference approach for causal effects in social networks with interference.
• To address unmeasured confounding at the cluster level in multilevel network data.
• To estimate the causal effect of the home environment on adolescent school performance.

Main Methods:

• A joint mixed-effects model for outcome and exposure combined with direct standardization.
• Bayesian inference framework to handle network interference and unmeasured cluster confounding.
• Simulation studies comparing the proposed method with traditional linear mixed and fixed effects models.

Main Results:

• The proposed Bayesian joint mixed-effects model achieves unbiased causal effect estimation.
• The method effectively handles network interference and unmeasured confounding in multilevel data.
• Simulations demonstrate superior performance compared to linear mixed and fixed effects models.

Conclusions:

• The developed Bayesian approach provides valid tools for causal effect estimation in complex social networks.
• This method is applicable to analyzing real-world data, such as the impact of home environment on adolescent academic outcomes.
• It offers a robust solution for unmeasured confounding and network interference in population studies.