A Graph-Theoretic Approach to Randomization Tests of Causal Effects under General Interference

Abstract Interference exists when a unit's outcome depends on another treatment assignment. For example, intensive policing one street could have spillover effect neighbouring streets. Classical randomization tests typically break down in this setting because many null hypotheses of interest are no longer sharp under interference. A promising alternative is to instead construct conditional test subset units and assignments for which given hypothesis sharp. Finding these subsets challenging, however, existing methods limited special cases or power. In paper, we propose valid easy-to-implement general class arbitrary interference between units. Our key idea represent the as bipartite graph assignments, find an appropriate biclique graph. Importantly, within biclique, enabling randomization-based tests. We also connect size statistical Moreover, can apply off-the-shelf clustering such bicliques efficiently at scale. illustrate our approach settings with clustered show advantages over designed specifically that setting. then method large-scale experiment Medellín, Colombia, where has spatial structure.