Models, methods and network topology: Experimental design for the study of interference

•The statistical power of network experiments depends on propagation models.•More precision about spillover arises when fewer units receive treatment.•More precision in general arises when node and treatment are not correlated.