Discussion on 'Causal inference with misspecified network interference structure' by Bar Weinstein and Daniel Nevo

Area of Science:

• Statistics
• Causal Inference
• Network Analysis

Background:

• Network misspecification is a common issue in statistical modeling, leading to biased estimations.
• Understanding interference structures is crucial for accurate causal inference in networked systems.

Purpose of the Study:

• To derive the bias of mean potential outcome estimators under network misspecification.
• To develop robust estimators that mitigate bias caused by incorrect network structures.
• To explore strategies for selecting appropriate network candidates for robust estimation.

Main Methods:

• Derivation of the bias formula for mean potential outcome estimators.
• Development of Network Misspecification Robust (NMR) estimators.
• Analysis of bias-variance tradeoffs and assumption relaxations for NMR estimators.

Main Results:

• Bias in estimation is directly related to the divergence between misspecified and true network structures.
• NMR estimators achieve unbiasedness when at least one candidate network accurately reflects the true interference structure.
• The study identifies open questions regarding the selection of candidate network pools for NMR estimators.

Conclusions:

• Network misspecification poses a significant challenge in causal inference, but robust estimation methods can be developed.
• NMR estimators offer a promising approach to unbiased estimation in the presence of potential network misspecification.
• Future research should focus on practical selection strategies for candidate networks and real-world applications.