Causal inference with misspecified network interference structure

Area of Science:

• Causal inference
• Network analysis
• Social network analysis

Background:

• Interference between units is common in many fields.
• Interference patterns are often modeled using networks.
• Accurate network specification is crucial but challenging.

Purpose of the Study:

• To investigate the consequences of network misspecification in causal effect estimation.
• To develop a novel estimator robust to network misspecification.

Main Methods:

• Derivation of bias bounds for misspecified networks.
• Quantification of bias using induced exposure probabilities.
• Development of a novel estimator leveraging multiple networks.

Main Results:

• Estimation bias increases with network divergence.
• The proposed estimator is unbiased if at least one network is correct.
• Simulations and field experiment demonstrate estimator utility.

Conclusions:

• Network misspecification poses a significant challenge in causal inference.
• The proposed multi-network estimator offers robustness to network specification errors.
• This approach enhances the reliability of causal effect estimation in networked settings.