Abstract

Dependency in large networks is among the most intractable challenges in social network analysis, as current approaches are computationally infeasible in networks with thousands of nodes. In this article, we introduce path-weighted regression (PWR), a novel and computationally efficient method to assess spatial dependency and non-stationarity in large networks. PWR estimates separate models for each node in the network and weighs nearby nodes more heavily than distant nodes. This approach allows us to estimate local effects in a large network efficiently. We illustrate PWR using a large Twitter network with 280,000 nodes showing the reaction by Republicans and Democrats to Trump’s Covid-19 diagnosis