🔎 The Problem

Dyadic data are widespread in the social sciences, but standard inference often breaks down because multiple dyads share members and therefore have correlated errors. This complex clustering structure is frequently ignored, producing unreliable standard errors and misleading conclusions.

🧾 What Was Introduced

• A non-parametric, sandwich-type robust variance estimator tailored for linear regression with dyadic dependence.
• A clear set of conditions under which the estimator is consistent, enabling principled inference in the presence of shared-membership clustering.

🛠️ How the Method Was Extended and Implemented

• Extensions cover repeated observations, weighted observations, directed dyads, and longitudinal (panel) dyadic data.
• Implementation is provided for generalized linear models, including logistic regression, broadening applicability beyond OLS.

🧪 Evidence on Performance

• Simulation studies assess finite-sample behavior and compare the estimator against alternatives.
• An empirical application to interstate disputes illustrates practical use and highlights differences in inference when dyadic clustering is accounted for.

📌 Why It Matters

Accounting for shared-member clustering in dyadic settings fixes a common source of inferential error. The proposed sandwich estimator and its extensions offer a practical route to more reliable standard errors across a range of dyadic research designs and model types.