A0842
Title: Permutation tests for dyadic models
Authors: Pujee Tuvaandorj - York University (Canada) [presenting]
Abstract: Dyadic data models are commonly employed to analyze interactions between population units, where it is essential to properly account for the cross-sectional dependence induced by the dyadic structure. Well-known examples include international trade, where each observation represents a pair of countries, and network formation, where binary variables indicate whether pairs of units share a link. Permutation tests are developed for generalized linear models with such data structure. The large sample validity of the proposed tests is established, conditions under which they remain exact in finite samples are characterized, and robustness in sparse dyads is examined. Simulations suggest better finite sample properties of the proposed permutation tests compared to the asymptotic tests. Applications to the gravity model of international trade are provided, along with a re-examination of the democratic trade hypothesis in international relations.
