This paper introduces a robust estimation framework for network panel data that handles unbalanced panel structures and outliers. It develops a unified M-estimation approach with Huber-type loss functions and a novel weighted within-transformation for unbalanced network structures, maintaining statistical efficiency and breakdown point guarantees.
Key findings
Unified M-estimation approach integrates Huber-type robust loss functions.
Novel weighted within-transformation accommodates fixed effects under arbitrary missing data patterns.
Consistency and asymptotic normality established under a double asymptotic framework.
Substantial improvements in bias, mean squared error, and coverage probabilities demonstrated through simulations.
Limitations & open questions
Further research needed for dynamic panels, heterogeneous peer effects, and nonlinear outcomes.